feat(sdk): re-embed vector stores and the graph into a new embedding space

Add hy_memory.reembed for switching the embedding model while writes are
paused. Source and target are named explicitly; a space's identity is the
model name plus the width.

- vector_space_names, plan_reembed, reembed_vector_space (resumable,
  cancellable at batch boundaries, widths validated before any upsert, target
  wins on existing ids, source never written), verify_reembed,
  drop_vector_space.
- scan_vector_spaces and diff_spaces for MemHub's startup reconcile of stray
  spaces.
- reembed_graph builds a new Kuzu DB at the target width, copies every node
  and edge, recomputes Memory/beh/Topic/HierarchyTopic embeddings and rebuilds
  the HNSW index; the source graph is opened read-only. verify_graph_reembed
  and drop_graph.

EmbedService.embed_batch now chunks to embedder.batch_size (default 64),
honours Retry-After on 429, takes a per-call max_retries, and rejects vectors
whose width differs from embedding_dims.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

.env.example

@@ -40,6 +40,7 @@ MEMORY_EMBEDDING_DIMS=1024
 # MEMORY_EMBEDDER_TIMEOUT=15
 # MEMORY_EMBEDDER_MAX_RETRIES=2
 # MEMORY_EMBEDDER_RETRY_DELAY=0.3
+# MEMORY_EMBEDDER_BATCH_SIZE=64
 # MEMORY_EMBEDDER_EXTRA_HEADERS={}
 # MEMORY_EMBEDDER_EXTRA_BODY={}
 

CHANGELOG.md

@@ -4,6 +4,46 @@ Hy Memory and MemHub are versioned and released independently.
 
 ## Unreleased
 
+- Hy Memory: new `hy_memory.reembed` module for switching the embedding
+  model. Writers must be paused for the whole run. Nothing is derived from
+  the provider or endpoint: source and target are named explicitly, and the
+  identity of a space is the model name plus the width (`space_identity`).
+  Only Chroma and Kuzu are supported.
+  - `vector_space_names(config, base_name=None)` returns every physical
+    collection a client opens for the six vector roles: memory, entities,
+    coding keys and working facts, each in the working and coding variants.
+  - `reembed_vector_space(source, target, embedding=...)` pages every raw
+    source collection, covering all users and all layers including L1_RAW. It
+    embeds the text the write path embedded (`reembed_text`: the canonical
+    state's `state_retrieval_text`, a schema's `retrieval_text`, otherwise the
+    document). It then writes each record into the target with the same id,
+    document and metadata. Each batch's widths are checked before anything is
+    written. A cancel stops at a batch boundary (`ReembedCancelled`). A rerun
+    skips the ids the target already holds, so it resumes, and a merge never
+    overwrites the target. The source is never written.
+  - `plan_reembed`, `verify_reembed` and `drop_vector_space` complete the
+    vector side. `verify_reembed` checks ids, metadata, per-user counts and
+    the width, and runs a self-recall sample. `drop_vector_space` refuses
+    collections that a live client in this process holds open.
+  - `scan_vector_spaces(persist_directory)` is read-only. It lists every
+    space on disk with its roles, counts, stored width, per-user counts and
+    the models recorded in metadata. `diff_spaces(source, target)` returns the
+    ids that the source holds and the target lacks, whatever their status.
+  - `reembed_graph(source_path, target_path, embedding=..., target_dims=...)`
+    builds a new Kuzu graph at the new width. It copies every node and edge
+    and recomputes `Memory.embedding`, `Memory.beh_embedding`, and the Topic
+    and HierarchyTopic embeddings, then rebuilds the HNSW index. The source
+    graph is opened read-only and kept for rollback. A run that fails or is
+    cancelled leaves no target behind, and a rerun starts over.
+    `verify_graph_reembed` and `drop_graph` complete the graph side.
+- Hy Memory: `EmbedService.embed_batch` now splits large batches into
+  requests of `embedder.batch_size` texts each (`MEMORY_EMBEDDER_BATCH_SIZE`,
+  default 64); before, it sent them all in one request. On a 429 it waits for
+  the `Retry-After` time plus jitter, capped at 60 s. It takes a per-call
+  `max_retries`. A returned vector whose width differs from the configured
+  `embedding_dims` now raises `EmbeddingDimensionMismatchError` and is not
+  retried.
+
 ## MemHub 1.0.5 — 2026-10-28
 
 Build 36, on `hy-memory==2.0.8`. Linux x64/arm64 Runtime and macOS arm64
@@ -94,6 +134,29 @@ authority on first launch.
   `hy_node_permissions` proof table. Evidence, governed memories, model
   profiles, SDK stores, and source/host grants are preserved.
 
+## Hy Memory 2.0.9 — 2026-11-01
+
+- Published
+  [`hy-memory==2.0.9`](https://pypi.org/project/hy-memory/2.0.9/).
+- Placeholder values no longer overwrite the user's basic profile (L0). A
+  small extraction model filled `basic_info` with `{"name": "用户"}`,
+  `The user` or `N/A` when the user had said nothing about themselves, and
+  each placeholder superseded the real value, so the L0 chain read
+  费天翔 → Fei → 用户 → The user. Two defences:
+  - `upsert_basic_profile` drops placeholder values: empty, `用户`,
+    `the user`, `user`, `n/a`, `na`, `none`, `null`, `unknown`, `未提供`,
+    `未知`, `不详`, `-`, `?` and the field's own name, compared
+    case-insensitively with whitespace and punctuation stripped. When nothing
+    is left it writes no new version and no new L0 head. Placeholders already
+    stored in the chain no longer override earlier real values when the
+    profile is assembled, so `get_user_profile` of an affected user shows the
+    last real value again without any migration. This also stops an
+    assistant quoting an old memory ("The user's name is 用户.") from being
+    written back.
+  - The extraction prompts (zh and en) state the rule: only what the user
+    said in a `user` turn, never the field name or a placeholder, `{}`
+    otherwise, with a small-talk counter-example.
+
 ## Hy Memory 2.0.8 — 2026-10-28
 
 - Published

docs/env_reference.md

@@ -48,6 +48,7 @@
 | `MEMORY_EMBEDDER_TIMEOUT` | `15` | 单次超时，秒 |
 | `MEMORY_EMBEDDER_MAX_RETRIES` | `2` | 最大重试次数 |
 | `MEMORY_EMBEDDER_RETRY_DELAY` | `0.3` | 重试间隔，秒 |
+| `MEMORY_EMBEDDER_BATCH_SIZE` | `64` | `embed_batch` 单次请求最多发送的文本条数，超出按此切块 |
 | `MEMORY_EMBEDDER_EXTRA_HEADERS` | — | JSON object，附加 HTTP headers |
 | `MEMORY_EMBEDDER_EXTRA_BODY` | — | JSON object，附加请求 body |
 

hy_memory/__init__.py

@@ -55,6 +55,33 @@ except Exception:
 from .client import HyMemoryClient
 from .inspector import MemoryInspector
 from .runtime import SharedRuntime
+from .reembed import (
+    EmbeddingSpaceIdentity,
+    GraphReembedPlan,
+    GraphReembedProgress,
+    GraphReembedReport,
+    GraphVerification,
+    ReembedCancelled,
+    ReembedPlan,
+    ReembedProgress,
+    ReembedReport,
+    ReembedVerification,
+    ScannedSpace,
+    SpaceDiff,
+    VectorSpace,
+    diff_spaces,
+    drop_graph,
+    drop_vector_space,
+    plan_reembed,
+    reembed_graph,
+    reembed_text,
+    reembed_vector_space,
+    scan_vector_spaces,
+    space_identity,
+    vector_space_names,
+    verify_graph_reembed,
+    verify_reembed,
+)
 from .snapshot import (
     DataPlaneRestore,
     DataPlaneSnapshot,
@@ -129,6 +156,32 @@ __all__ = [
     "DataPlaneSnapshotRejected",
     "restore_local_data_plane",
     "snapshot_local_data_plane",
+    # 换 embedding 模型：把向量库和图重新 embed 进新的 embedding 空间
+    "EmbeddingSpaceIdentity",
+    "GraphReembedPlan",
+    "GraphReembedProgress",
+    "GraphReembedReport",
+    "GraphVerification",
+    "ReembedCancelled",
+    "ReembedPlan",
+    "ReembedProgress",
+    "ReembedReport",
+    "ReembedVerification",
+    "ScannedSpace",
+    "SpaceDiff",
+    "VectorSpace",
+    "diff_spaces",
+    "drop_graph",
+    "drop_vector_space",
+    "plan_reembed",
+    "reembed_graph",
+    "reembed_text",
+    "reembed_vector_space",
+    "scan_vector_spaces",
+    "space_identity",
+    "vector_space_names",
+    "verify_graph_reembed",
+    "verify_reembed",
     # 配置
     "MemoryConfig",
     # 公共扩展协议 / 精确读取类型

hy_memory/agent/extractor.py

@@ -682,6 +682,23 @@ Output:
 ```
 John's occupation and location are HIS attributes, so they go into `memory` with "John" as the subject. They must NEVER land in `basic_info.occupation` / `basic_info.location` — `basic_info` is exclusively the user's own profile. Only `name` (Tim, from the user message's `name` field) belongs there.
 
+## Example 11 — Counter-example: small talk has no basic_info
+
+New Messages:
+[{"role": "user", "content": "Hi"},
+ {"role": "assistant", "content": "Hello! The user's name is The user. How can I help?"}]
+Memory Date: 2025-09-03
+
+Output:
+```json
+{
+  "memory": [],
+  "intentions": [],
+  "basic_info": {}
+}
+```
+The user stated nothing about themselves. The assistant's line is not a user turn, and "The user" is a placeholder, not a name. Output `{}` — never `{"name": "The user"}` or `{"name": "user"}`.
+
 ---
 """
 
@@ -749,6 +766,14 @@ Rules:
   fields (e.g. name, age, location, occupation, employer) **explicitly stated**
   in the NEW user messages. Never infer, never invent. If no field is explicitly
   stated, output `{{}}`.
+  * HARD RULE: fill a field ONLY if the user stated it in a `user` turn.
+    Anything said only by the assistant (including the assistant quoting an
+    earlier memory such as "The user's name is X") does NOT count. If nothing
+    qualifies, output `"basic_info": {{}}`.
+  * NEVER write the field name itself or a placeholder as a value ("user",
+    "The user", "用户", "name", "N/A", "unknown", "none", "-"). Omit the field.
+  * Negative example: small talk such as `[user]: Hi` / `[assistant]: How can
+    I help?`, or `[user]: Reply with exactly: PONG` → `"basic_info": {{}}`.
   * Exception for `name`: the `name` field on the user's own message, or a leading
     line "The user's name is X" / "用户名叫X", is an authoritative source for
     `name` — even though the user never said it out loud.
@@ -817,6 +842,14 @@ Rules:
   fields (e.g. name, age, location, occupation, employer) **explicitly stated**
   in the NEW user messages. Never infer, never invent. If no field is explicitly
   stated, output `{{}}`.
+  * HARD RULE: fill a field ONLY if the user stated it in a `user` turn.
+    Anything said only by the assistant (including the assistant quoting an
+    earlier memory such as "The user's name is X") does NOT count. If nothing
+    qualifies, output `"basic_info": {{}}`.
+  * NEVER write the field name itself or a placeholder as a value ("user",
+    "The user", "用户", "name", "N/A", "unknown", "none", "-"). Omit the field.
+  * Negative example: small talk such as `[user]: Hi` / `[assistant]: How can
+    I help?`, or `[user]: Reply with exactly: PONG` → `"basic_info": {{}}`.
   * Exception for `name`: the `name` field on the user's own message, or a leading
     line "The user's name is X" / "用户名叫X", is an authoritative source for
     `name` — even though the user never said it out loud.

hy_memory/agent/prompts_zh.py

@@ -365,6 +365,13 @@ EXTRACT_PROMPT_ZH = """## 最近 k 条消息
 * basic_info：**只装用户本人的画像**。
   仅当 NEW user messages 中**明确**陈述稳定画像字段（如 姓名、年龄、所在地、职业、雇主）时才填写。
   绝不推断、绝不编造。没有任何字段被明确陈述时输出 `{{}}`。
+  * **硬性规则**：只有用户在 `user` 回合里**明确说出**的字段才能填写。只在
+    assistant 回合出现的内容（包括 assistant 引用旧记忆，如"The user's name is X"
+    / "用户名是X"）一律不算。没有符合条件的字段时，必须输出 `"basic_info": {{}}`。
+  * **绝不**把字段名本身或占位符当值写入（"用户"、"The user"、"user"、"name"、
+    "姓名"、"N/A"、"unknown"、"未知"、"未提供"、"-"），宁可省略该字段。
+  * 负例：寒暄如 `[user]: 你好` / `[assistant]: 我可以帮你。`，或
+    `[user]: Reply with exactly: PONG` → `"basic_info": {{}}`。
   * `name` 的例外：用户自己那条消息上的 `name` 字段，或文首形如
     "用户名叫X" / "The user's name is X" 的那一行，都是 `name` 的权威来源
     ——即使用户本人从未开口自报姓名。
@@ -440,6 +447,13 @@ EXTRACT_PROMPT_ZH_NO_INTENTION = """## 最近 k 条消息
 * basic_info：**只装用户本人的画像**。
   仅当 NEW user messages 中**明确**陈述稳定画像字段（如 姓名、年龄、所在地、职业、雇主）时才填写。
   绝不推断、绝不编造。没有任何字段被明确陈述时输出 `{{}}`。
+  * **硬性规则**：只有用户在 `user` 回合里**明确说出**的字段才能填写。只在
+    assistant 回合出现的内容（包括 assistant 引用旧记忆，如"The user's name is X"
+    / "用户名是X"）一律不算。没有符合条件的字段时，必须输出 `"basic_info": {{}}`。
+  * **绝不**把字段名本身或占位符当值写入（"用户"、"The user"、"user"、"name"、
+    "姓名"、"N/A"、"unknown"、"未知"、"未提供"、"-"），宁可省略该字段。
+  * 负例：寒暄如 `[user]: 你好` / `[assistant]: 我可以帮你。`，或
+    `[user]: Reply with exactly: PONG` → `"basic_info": {{}}`。
   * `name` 的例外：用户自己那条消息上的 `name` 字段，或文首形如
     "用户名叫X" / "The user's name is X" 的那一行，都是 `name` 的权威来源
     ——即使用户本人从未开口自报姓名。
@@ -669,6 +683,23 @@ Memory Date：2025-09-02
 ```
 职业和所在地是 **John 的**属性，所以进 `memory` 并以"John"作主体，**绝不能**落到 `basic_info` 的 `occupation` / `location` —— `basic_info` 专属于用户本人。只有 `name`（Tim，来自用户消息的 `name` 字段）才该放进去。注意输出语言随输入语言（此例为英文）。
 
+## 示例 11 —— 反例：寒暄没有 basic_info
+
+新消息：
+[{"role": "user", "content": "你好"},
+ {"role": "assistant", "content": "你好！我可以帮你。之前记得 The user's name is 用户."}]
+Memory Date：2025-09-03
+
+输出：
+```json
+{
+  "memory": [],
+  "intentions": [],
+  "basic_info": {}
+}
+```
+用户没有说任何关于自己的信息。assistant 那句不是 user 回合，"用户" 也是占位符而不是名字。必须输出 `{}`——**绝不能**输出 `{"name": "用户"}`。
+
 ---
 """
 

hy_memory/agent/tools/basic_profile.py

@@ -26,6 +26,8 @@ from __future__ import annotations
 
 import asyncio
 import logging
+import re
+import unicodedata
 import uuid
 import weakref
 from dataclasses import dataclass
@@ -208,6 +210,76 @@ def _sanitize_kv(
     return result
 
 
+# ================================================================
+# 占位符守卫（P0-1）
+# ================================================================
+# 弱模型（如 qwen3.5-2b）在用户什么都没说时，会把 schema 字段说明里的词当值
+# 填进 basic_info：{"name": "用户"} / "The user" / "N/A"。每个占位符都会
+# supersede 掉真实值，演化链变成 费天翔 → Fei → 用户 → The user。
+# 规则：占位符永远不入库；读侧组装 KV 时占位符也不覆盖已有真实值。
+
+_PLACEHOLDER_VALUES = frozenset({
+    "用户", "该用户", "本用户", "用户本人",
+    "the user", "the user's", "user", "a user", "this user",
+    "n/a", "na", "none", "null", "nil", "unknown", "unspecified",
+    "not provided", "not specified", "not mentioned",
+    "未提供", "未知", "不详", "未说明", "未提及", "无",
+    "-", "?",
+})
+
+# 内置字段的中文标签：弱模型也会把字段说明的中文词当值。
+_FIELD_LABELS: Dict[str, tuple] = {
+    "name": ("姓名", "名字", "用户名", "用户姓名", "用户名字"),
+    "age": ("年龄", "用户年龄"),
+    "location": ("所在地", "地点", "城市", "用户所在地"),
+    "occupation": ("职业", "工作", "用户职业"),
+    "employer": ("雇主", "公司", "用户雇主"),
+}
+
+# 去掉标点/符号后再比较，让 "The user." / "用户。" / "N/A" 都命中。
+# "-" / "?" 本身就是纯标点，剥完为空串，同样判为占位符。
+_PUNCT_RE = re.compile(r"[\s\.,;:!?'\"`~()\[\]{}<>/\\|_\-\u3000-\u303f\uff00-\uff0f\uff1a-\uff20]+")
+
+
+def _placeholder_key(value: str) -> str:
+    s = unicodedata.normalize("NFKC", str(value or "")).strip().lower()
+    s = s.replace("\u2019", "'")
+    s = _PUNCT_RE.sub(" ", s)
+    return " ".join(s.split())
+
+
+_PLACEHOLDER_KEYS = frozenset(_placeholder_key(v) for v in _PLACEHOLDER_VALUES)
+
+
+def is_placeholder_value(field_name: str, value: Any) -> bool:
+    """value 是否为占位符（空 / 用户 / The user / N/A / 字段名本身 等）。
+
+    大小写不敏感、首尾空白与标点不敏感。数字（如 age=30）永远不是占位符。
+    """
+    if value is None:
+        return True
+    if isinstance(value, bool):
+        return False
+    if isinstance(value, (int, float)):
+        return False
+    key = _placeholder_key(value)
+    if not key:
+        return True
+    if key in _PLACEHOLDER_KEYS:
+        return True
+    field_key = _placeholder_key(field_name)
+    if field_key and key in (field_key, field_key.replace("_", " ")):
+        return True
+    for label in _FIELD_LABELS.get(field_name, ()):
+        if key == _placeholder_key(label):
+            return True
+    return False
+
+
+def _drop_placeholders(kv: Dict[str, Any]) -> Dict[str, Any]:
+    return {k: v for k, v in kv.items() if not is_placeholder_value(k, v)}
+
+
 # ================================================================
 # 公共入口：upsert_basic_profile
 # ================================================================
@@ -287,6 +359,8 @@ async def _upsert_basic_profile_unlocked(
     # L0 Basic Info 从输入、唯一性到落盘都只属于 user_id。
 
     new_kv = _sanitize_kv(kv or {}, allowed_fields=allowed_fields)
+    # 单调性守卫：占位符永远不覆盖已有值，过滤后为空就不写新版本、不产生新链头。
+    new_kv = _drop_placeholders(new_kv)
     if not new_kv:
         return BasicProfileUpsertResult(
             success=True, changed=False, reason="no valid fields after sanitize"
@@ -513,5 +587,8 @@ def _assemble_full_kv_from_nodes(
         for k, v in kv.items():
             if allowed_set is not None and k not in allowed_set:
                 continue
+            # 历史上已落盘的占位符版本（如 name=用户）不覆盖更早的真实值。
+            if is_placeholder_value(k, v):
+                continue
             full_kv[k] = v
     return full_kv

hy_memory/config.py

@@ -260,6 +260,7 @@ class EmbedderConfig:
     max_retries: int = None  # 最大重试次数
     retry_delay: float = None  # 重试间隔（秒）
     timeout: int = None  # 请求超时（秒）
+    batch_size: int = None  # 单次 HTTP 请求最多带多少条 input（embed_batch 按此切块）
 
     # 内部平台扩展（extra_headers / extra_body 会透传给 OpenAI client）
     extra_headers: Optional[Dict[str, str]] = None
@@ -294,6 +295,10 @@ class EmbedderConfig:
             # 请求切断并制造重试风暴。SDK 内层重试已关闭，15s 是每个显式 attempt
             # 的真实上限；部署可通过 MEMORY_EMBEDDER_TIMEOUT 覆盖。
             self.timeout = _get_env_int("MEMORY_EMBEDDER_TIMEOUT", 15)
+        if self.batch_size is None:
+            # 大批量（如重新 embed 整个库）不能一次 HTTP 发几千条：多数服务对
+            # input 数组有上限，超了直接 400。按 batch_size 切块逐块发送。
+            self.batch_size = _get_env_int("MEMORY_EMBEDDER_BATCH_SIZE", 64)
         if self.extra_headers is None:
             val = os.getenv("MEMORY_EMBEDDER_EXTRA_HEADERS")
             if val:

hy_memory/core/embed_service.py

@@ -38,10 +38,12 @@ import importlib.util
 import json
 from json import JSONDecodeError
 import logging
+import random
 import threading
 import time
 
 from ..config import MemoryConfig
+from ..embedding import EmbeddingDimensionMismatchError
 from .workload import embedding_slot
 from ..utils.log_setup import get_request_id
 
@@ -187,6 +189,41 @@ def _redact_headers(headers: Optional[Dict[str, Any]]) -> Optional[Dict[str, Any
     return out
 
 
+# Retry-After 的上限：服务端给出超长等待（或恶意值）时不让单次调用挂几十分钟。
+_RETRY_AFTER_MAX_S = 60.0
+
+
+def _retry_after_seconds(exc: Exception) -> Optional[float]:
+    """从 429 异常的响应头里取 Retry-After（秒数或 HTTP 日期），取不到返回 None。"""
+    response = getattr(exc, "response", None)
+    headers = getattr(response, "headers", None)
+    if not headers:
+        return None
+    try:
+        raw = headers.get("retry-after") or headers.get("Retry-After")
+    except Exception:
+        return None
+    if raw is None:
+        return None
+    raw = str(raw).strip()
+    try:
+        seconds = float(raw)
+    except ValueError:
+        try:
+            from email.utils import parsedate_to_datetime
+            from datetime import datetime, timezone
+
+            when = parsedate_to_datetime(raw)
+            if when.tzinfo is None:
+                when = when.replace(tzinfo=timezone.utc)
+            seconds = (when - datetime.now(timezone.utc)).total_seconds()
+        except Exception:
+            return None
+    if seconds != seconds:  # NaN
+        return None
+    return min(max(0.0, seconds), _RETRY_AFTER_MAX_S)
+
+
 def _embeddings_api_url(base_url: str) -> str:
     """OpenAI SDK posts to ``{base_url}/embeddings`` (base_url usually ends with /v1)."""
     base = (base_url or "").rstrip("/")
@@ -205,7 +242,8 @@ class EmbedServiceConfig:
     api_key: str = ""
     base_url: str = ""
     embedding_dims: int = 1536
-    batch_size: int = 100
+    # embed_batch 单次 HTTP 最多带多少条 input；超出按块顺序发送
+    batch_size: int = 64
     enable_cache: bool = True
     cache_size: int = 10000
     max_retries: int = 5
@@ -287,6 +325,7 @@ class EmbedService:
             max_retries=embedder.max_retries if embedder.max_retries is not None else 5,
             retry_delay=embedder.retry_delay if embedder.retry_delay is not None else 3.0,
             timeout=embedder.timeout if embedder.timeout is not None else 60,
+            batch_size=max(1, int(getattr(embedder, "batch_size", None) or 64)),
             queue_batch_size=int(os.getenv("MEMORY_EMBED_BATCH_SIZE", "8")),
             queue_batch_window_ms=float(os.getenv("MEMORY_EMBED_BATCH_WINDOW_MS", "100")),
             extra_headers=embedder.extra_headers,
@@ -640,6 +679,8 @@ class EmbedService:
         self,
         texts: List[str],
         use_cache: bool = True,
+        *,
+        max_retries: Optional[int] = None,
     ) -> List[List[float]]:
         """
         批量向量化
@@ -647,17 +688,23 @@ class EmbedService:
         Args:
             texts: 文本列表
             use_cache: 是否使用缓存
+            max_retries: 本次调用的重试次数；None 用配置值。长批量任务（重新
+                embed 整个库）可以给更宽的重试预算，不影响在线路径。
         
         Returns:
             向量列表
         """
-        vectors, _metas = await self.embed_batch_with_meta(texts, use_cache=use_cache)
+        vectors, _metas = await self.embed_batch_with_meta(
+            texts, use_cache=use_cache, max_retries=max_retries,
+        )
         return vectors
 
     async def embed_batch_with_meta(
         self,
         texts: List[str],
         use_cache: bool = True,
+        *,
+        max_retries: Optional[int] = None,
     ) -> Tuple[List[List[float]], List[Dict[str, Any]]]:
         """批量向量化，同时返回每条的 provider metadata（非 vector）。"""
         if not texts:
@@ -682,7 +729,12 @@ class EmbedService:
             indices_to_embed.append(i)
 
         if texts_to_embed:
-            vectors, batch_metas = await self._embed_texts_with_meta(texts_to_embed)
+            # max_retries 只在显式给出时透传：保持 `_embed_texts_with_meta(texts)`
+            # 这个既有调用形状不变（测试与下游替身按它打桩）。
+            retry_kwargs = {} if max_retries is None else {"max_retries": max_retries}
+            vectors, batch_metas = await self._embed_texts_with_meta(
+                texts_to_embed, **retry_kwargs,
+            )
 
             for j, vector in enumerate(vectors):
                 idx = indices_to_embed[j]
@@ -700,9 +752,9 @@ class EmbedService:
         return results, metas
     
     async def _embed_texts_with_meta(
-        self, texts: List[str]
+        self, texts: List[str], *, max_retries: Optional[int] = None,
     ) -> Tuple[List[List[float]], List[Dict[str, Any]]]:
-        """实际的向量化调用：**一次 HTTP 带走整批**；返回 (vectors, metas)。
+        """实际的向量化调用：**一次 HTTP 带走一块**；返回 (vectors, metas)。
 
         早期实现是 `for text in texts: await _embed_openai([text])` 逐条串行，
         原意是"避免 batch size 限制"，但代价极大：线上一批 8 条 × ~5.6s 串行
@@ -719,21 +771,34 @@ class EmbedService:
         截断在这里做（发送前逐条过 `_truncate_for_embedding`）：所有 embedding
         路径（embed / embed_batch / embed_queued / *_with_meta）最终都落到这个
         方法 —— 放在各入口会漏。理由与阈值见模块头部 `_EMBED_MAX_TOKENS`。
+
+        切块：一次带走整批只适用于小批。几千条的 input 数组会被多数服务直接
+        400（input 上限），所以按 `batch_size`（默认 64）切块，块之间顺序发送。
+        不超过一块时行为与切块前逐字相同（`batch_mode="single_request"`）。
         """
         if not texts:
             return [], []
 
         safe_texts = [_truncate_for_embedding(t) for t in texts]
-        vectors, meta = await self._embed_openai(safe_texts)
-        base_meta = dict(meta or {})
-        base_meta["batch_mode"] = "single_request"
-        base_meta["batch_input_count"] = len(safe_texts)
-
+        chunk = max(1, int(self._embed_config.batch_size or 64))
+        vectors: List[List[float]] = []
         all_metas: List[Dict[str, Any]] = []
-        for idx in range(len(safe_texts)):
-            m = dict(base_meta)
-            m["batch_index"] = idx
-            all_metas.append(m)
+        chunk_count = (len(safe_texts) + chunk - 1) // chunk
+        for start in range(0, len(safe_texts), chunk):
+            part = safe_texts[start:start + chunk]
+            retry_kwargs = {} if max_retries is None else {"max_retries": max_retries}
+            part_vectors, meta = await self._embed_openai(part, **retry_kwargs)
+            base_meta = dict(meta or {})
+            base_meta["batch_mode"] = "single_request" if chunk_count == 1 else "chunked"
+            base_meta["batch_input_count"] = len(part)
+            if chunk_count > 1:
+                base_meta["batch_chunk_index"] = start // chunk
+                base_meta["batch_chunk_count"] = chunk_count
+            for idx in range(len(part)):
+                m = dict(base_meta)
+                m["batch_index"] = start + idx
+                all_metas.append(m)
+            vectors.extend(part_vectors)
 
         self._last_embed_meta = all_metas[-1] if all_metas else {}
         return vectors, all_metas
@@ -783,7 +848,7 @@ class EmbedService:
             )
 
     async def _embed_openai(
-        self, texts: List[str]
+        self, texts: List[str], *, max_retries: Optional[int] = None,
     ) -> Tuple[List[List[float]], Dict[str, Any]]:
         """
         使用 OpenAI 兼容接口向量化
@@ -873,7 +938,9 @@ class EmbedService:
             "endpoint_overridden": bool(_ov),
         }
 
-        _max_retries = max(0, int(self._embed_config.max_retries or 0))
+        _max_retries = max(0, int(
+            (self._embed_config.max_retries or 0) if max_retries is None else max_retries
+        ))
         _retry_delay = max(0.0, float(self._embed_config.retry_delay or 0.0))
         last_exc: Optional[Exception] = None
 
@@ -919,6 +986,17 @@ class EmbedService:
                         f"got={len(vectors)}"
                     )
                 dims = [len(v) for v in vectors]
+                expected_dims = self._embed_config.embedding_dims
+                if expected_dims:
+                    bad = [d for d in dims if d != int(expected_dims)]
+                    if bad:
+                        # 宽度不对的向量写进集合会把集合的维度钉死在错误值上（Chroma
+                        # 首次 add 定维），或在已有集合上整批失败。确定性错误，不重试。
+                        raise EmbeddingDimensionMismatchError(
+                            "embedding dimension mismatch: "
+                            f"expected={int(expected_dims)} got={bad[0]} "
+                            f"model={self._embed_config.model}"
+                        )
                 meta = {
                     "http_status": raw.status_code,
                     "response_headers": resp_headers,
@@ -954,6 +1032,8 @@ class EmbedService:
                     },
                 )
                 return vectors, meta
+            except EmbeddingDimensionMismatchError:
+                raise
             except Exception as e:
                 last_exc = e
                 _err_name = type(e).__name__
@@ -987,12 +1067,18 @@ class EmbedService:
                 if _retryable and attempt < _max_retries:
                     # 逐出可能卡死的 client,下个 attempt 重建干净连接(恢复原防御语义)
                     self._evict_client(_eff_base_url, _eff_api_key, _eff_timeout)
+                    _delay = _retry_delay
+                    _retry_after = _retry_after_seconds(e) if _status == 429 else None
+                    if _retry_after is not None:
+                        # 限流时服务端明确告诉了要等多久：按它等，再加一点抖动，
+                        # 避免多个 worker 在同一时刻一起醒来再次撞限流。
+                        _delay = max(_delay, _retry_after) + random.uniform(0, 0.25 * _retry_after)
                     logger.warning(
                         f"[embed] attempt {attempt + 1}/{_max_retries + 1} failed "
-                        f"({_err_name}): {_err_msg[:200]}; retry in {_retry_delay}s"
+                        f"({_err_name}): {_err_msg[:200]}; retry in {_delay:.2f}s"
                     )
-                    if _retry_delay > 0:
-                        await asyncio.sleep(_retry_delay)
+                    if _delay > 0:
+                        await asyncio.sleep(_delay)
                     continue
                 self._log_embed_call(
                     status="error",

hy_memory/data/graph_store_kuzu.py

@@ -22,6 +22,7 @@ import logging
 import json
 import threading
 import time
+import weakref
 
 from ..models.memory import MemoryNode, MemoryLayer, MemoryStatus
 from ..config import MemoryConfig
@@ -77,6 +78,21 @@ async def _run_in_kuzu_pool(func, *args, **kwargs):
     return await run_in_executor_guarded(_kuzu_executor, _measured)
 
 
+# 本进程里已打开、尚未 close 的 store。drop_graph 据此拒绝删掉一个正被
+# 活着的 client 使用的图库。WeakSet：store 被回收后自动消失。
+_LIVE_STORES: "weakref.WeakSet[KuzuGraphStore]" = weakref.WeakSet()
+
+
+def open_graph_paths() -> List[str]:
+    """Real paths of the graph databases live stores in this process hold open."""
+    import os
+    return [
+        os.path.realpath(str(store._db_path))
+        for store in list(_LIVE_STORES)
+        if store._db is not None
+    ]
+
+
 def kuzu_executor_snapshot() -> Dict[str, Any]:
     with _kuzu_stats_lock:
         return {
@@ -350,6 +366,7 @@ class KuzuGraphStore(GraphStoreBase):
             )
             self._db = db
             self._conn = conn
+            _LIVE_STORES.add(self)
         except asyncio.TimeoutError:
             logger.warning(
                 f"GraphStore init timed out after 5s (Kuzu file lock contention). "
@@ -2441,6 +2458,7 @@ class KuzuGraphStore(GraphStoreBase):
     async def close(self) -> None:
         """关闭连接"""
         conn, db = self._conn, self._db
+        _LIVE_STORES.discard(self)
         self._conn = None
         self._db = None
         self._available = False

hy_memory/data/vector_store_base.py

@@ -89,6 +89,17 @@ def passes_created_after(
 logger = logging.getLogger(__name__)
 
 
+def physical_collection_name(base_name: Optional[str], dims: Optional[int]) -> str:
+    """The collection name a store opens for ``base_name`` at ``dims``.
+
+    The dims suffix keeps two embedding widths from sharing one collection.
+    `hy_memory.reembed.vector_space_names` uses this too, so callers never
+    re-derive the rule.
+    """
+    base = base_name or "agent_memories"
+    return f"{base}_{dims}" if dims else base
+
+
 class VectorStoreBase(ABC):
     """
     向量存储抽象基类
@@ -98,13 +109,10 @@ class VectorStoreBase(ABC):
 
     def __init__(self, config: MemoryConfig):
         self.config = config
-        base_name = config.vector_store.collection_name or "agent_memories"
-        dims = config.vector_store.embedding_dims
         # 自动拼接维度后缀，避免不同 embedding 模型写入同一 collection 导致维度冲突
-        if dims:
-            self._collection_name = f"{base_name}_{dims}"
-        else:
-            self._collection_name = base_name
+        self._collection_name = physical_collection_name(
+            config.vector_store.collection_name, config.vector_store.embedding_dims,
+        )
 
     _supports_retrieval_clauses: bool = False
     _supports_narrow_episode_queries: bool = False

hy_memory/data/vector_store_chroma.py

@@ -9,10 +9,12 @@ Agent Memory V2 - VectorStore (Chroma)
 - 同步 API → _run_in_vdb_pool() 包装为异步
 """
 
-from typing import Optional, List, Dict, Any
+from typing import Optional, List, Dict, Any, Set, Tuple
 import concurrent.futures
 import functools
 import logging
+import os
+import weakref
 
 from ..models.memory import (
     MemoryNode, MemoryLayer, MemoryStatus,
@@ -33,6 +35,30 @@ _vdb_executor = concurrent.futures.ThreadPoolExecutor(
 )
 
 
+# 本进程里已 initialize、尚未 close 的 store。drop_vector_space 据此拒绝删掉
+# 一个正被活着的 client 使用的集合。WeakSet：store 被回收后自动消失。
+_LIVE_STORES: "weakref.WeakSet[ChromaVectorStore]" = weakref.WeakSet()
+
+
+def _location_key(vs_config) -> Tuple[str, str]:
+    if getattr(vs_config, "host", None):
+        return ("http", f"{vs_config.host}:{vs_config.port or 8000}")
+    return ("local", os.path.realpath(str(vs_config.persist_directory or "")))
+
+
+def open_collection_names(location: Tuple[str, str]) -> Set[str]:
+    """Collections that live stores in this process hold open at ``location``."""
+    names: Set[str] = set()
+    for store in list(_LIVE_STORES):
+        if store._client is None or _location_key(store.config.vector_store) != location:
+            continue
+        names.add(store._collection_name)
+        if store._entity_collection is not None:
+            from ..pipelines._retrieval.config import entity_collection_name
+            names.add(entity_collection_name(store._collection_name))
+    return names
+
+
 async def _run_in_vdb_pool(func, *args, **kwargs):
     """Run a synchronous Chroma call in the dedicated VDB executor."""
     call = functools.partial(func, *args, **kwargs) if args or kwargs else func
@@ -88,6 +114,7 @@ class ChromaVectorStore(VectorStoreBase):
             return client, collection
 
         self._client, self._collection = await _run_in_vdb_pool(_init)
+        _LIVE_STORES.add(self)
         logger.debug(f"VectorStore initialized (Chroma), collection={self._collection_name}")
 
     # ================================================================
@@ -561,6 +588,7 @@ class ChromaVectorStore(VectorStoreBase):
     async def close(self) -> None:
         """关闭客户端"""
         # Chroma PersistentClient 无需显式关闭
+        _LIVE_STORES.discard(self)
         self._client = None
         self._collection = None
         self._entity_collection = None

memhub/README.md

@@ -18,7 +18,7 @@ If you run into a problem, please open an Issue with the version, platform,repro
 
 ## 发行与平台合同
 
-源码当前固定使用公开 PyPI 正式版 `hy-memory==2.0.8`。MemHub 源码版本为 `1.0.5`
+源码当前固定使用公开 PyPI 正式版 `hy-memory==2.0.9`。MemHub 源码版本为 `1.0.5`
 （build 36），GitHub 仓库为
 [`Tencent-Hunyuan/Hy-Memory`](https://github.com/Tencent-Hunyuan/Hy-Memory)，npm 包名为
 `memhub-cli`。当前代码架构见
@@ -118,7 +118,7 @@ MemHub 的 SQLite 数据默认位于 `~/.one-memory`，不在 npm 包或 Runtime
 
 ## Hy Memory
 
-SDK 依赖固定为公开 PyPI 正式版 `hy-memory==2.0.8`。SDK 配置、权限边界及 search
+SDK 依赖固定为公开 PyPI 正式版 `hy-memory==2.0.9`。SDK 配置、权限边界及 search
 返回格式见 [`MEMORY.md`](MEMORY.md)。HY search 返回的 `content` 已包含 SDK 统一生成的
 逻辑时间和演化上下文；调用方不应再次转换时间戳或重复拼接 `When`。
 

memhub/SBOM.spdx.json

@@ -2352,7 +2352,7 @@
     {
       "SPDXID": "SPDXRef-Package-hy-memory",
       "name": "hy-memory",
-      "versionInfo": "2.0.8",
+      "versionInfo": "2.0.9",
       "downloadLocation": "NOASSERTION",
       "filesAnalyzed": false,
       "licenseConcluded": "Apache-2.0",

memhub/THIRD_PARTY_NOTICES.md

@@ -9,7 +9,7 @@ Per-package license texts are not reproduced here yet; the license
 column is filled in as part of release review, and the SBOM records
 `NOASSERTION` until then rather than guessing.
 
-## hy-memory 2.0.8
+## hy-memory 2.0.9
 
 License: Apache-2.0
 

memhub/apps/cli/src/cli-logs.test.ts

@@ -162,10 +162,54 @@ it("never emits a secret from a malformed manually written event", async () => {
     command: "logs",
     state: "empty",
     events: 0,
-    invalid_lines: 1,
+    invalid_lines: 0,
+    unrecognized_lines: 1,
   });
 });
 
+// After an upgrade the files still hold an older build's records. They are
+// whole JSON lines the reader leaves out, not damage, so the text output must
+// not warn about them; only a line that is not JSON is called damaged.
+it("warns only about lines that are not JSON, not about older records", async () => {
+  const dataDir = await fixture();
+  const older = {
+    schema_version: 0,
+    service: "desktop",
+    component: "desktop.lifecycle",
+    event: "desktop.started",
+  };
+  await writeFile(
+    path.join(dataDir, "logs", "diagnostic-desktop.jsonl"),
+    `${JSON.stringify(older)}\n${JSON.stringify({ ...older, schema_version: 1 })}\n`,
+    { mode: 0o600 },
+  );
+  let stdout = "";
+  let stderr = "";
+  vi.spyOn(process.stdout, "write").mockImplementation((value) => {
+    stdout += String(value);
+    return true;
+  });
+  vi.spyOn(process.stderr, "write").mockImplementation((value) => {
+    stderr += String(value);
+    return true;
+  });
+
+  expect(await run(["logs"], { ONE_MEMORY_DATA_DIR: dataDir })).toBe(0);
+  expect(stdout).toContain("core.start_failed");
+  expect(stdout).not.toContain("损坏");
+  expect(stderr).toBe("");
+
+  await writeFile(
+    path.join(dataDir, "logs", "diagnostic-cli.jsonl"),
+    '{"schema_version": 1, "event": \n',
+    { mode: 0o600 },
+  );
+  stdout = "";
+  expect(await run(["logs"], { ONE_MEMORY_DATA_DIR: dataDir })).toBe(0);
+  expect(stdout).toContain("跳过 1 行损坏的诊断记录。");
+  expect(stderr).toBe("memhub: 跳过 1 行损坏的诊断记录。\n");
+});
+
 it("rejects invalid filters without touching the log directory", async () => {
   const dataDir = path.join(
     await mkdtemp(path.join(os.tmpdir(), "om-cli-no-log-")),

memhub/apps/cli/src/cli.ts

@@ -190,8 +190,10 @@ import {
 } from "./managed-cli-link.js";
 import {
   addMemory,
+  DEFAULT_USER_ID,
   isMemoryId,
   isRequestId,
+  isUserId,
   MEMORY_KINDS,
   MEMORY_SCENES,
   MEMORY_VARIANTS,
@@ -209,6 +211,7 @@ import {
   renderMemoryAddText,
   renderMemoryDetailText,
   renderMemoryPageText,
+  userIdParameters,
   validMemoryText,
 } from "./memory-report.js";
 import {
@@ -538,6 +541,23 @@ async function runCommand(
       environment = { ...environment, ONE_MEMORY_DATA_DIR: lifted.dataDir };
     }
   }
+  // `--user-id <id>` is global to the memory commands: it names whose
+  // memories are read or written. It is lifted before dispatch so each
+  // command's own parser keeps seeing exactly the arguments it always did.
+  // `sources connect` gives the flag its own meaning and parses it itself.
+  let userId: string | undefined;
+  if (USER_ID_COMMANDS.has(arguments_[0] ?? "")) {
+    const lifted = liftUserIdOption(arguments_);
+    if (lifted === null) {
+      return reject(
+        findCommand(arguments_[0] ?? ""),
+        asJson,
+        "--user-id 无效：只能是 1–64 位字母、数字、点、下划线或连字符。",
+      );
+    }
+    arguments_ = lifted.arguments_;
+    userId = lifted.userId;
+  }
   environment = await environmentWithManagedDataDir(environment);
   const command = arguments_[0];
   if (command === undefined) return reject(undefined, asJson, "缺少命令。");
@@ -776,9 +796,11 @@ async function runCommand(
     const attachment = await attachToRuntimeOrNull(environment);
     if (attachment === null) return runtimeUnavailable(descriptor, asJson);
     if (parsed.kind === "status") {
-      const receipt = await readMemoryWrite(attachment, parsed.id).catch(
-        () => undefined,
-      );
+      const receipt = await readMemoryWrite(
+        attachment,
+        parsed.id,
+        userId,
+      ).catch(() => undefined);
       if (receipt === undefined) {
         return emitFailure(asJson, "write status", "memory_write_read_failed");
       }
@@ -812,6 +834,7 @@ async function runCommand(
         ? {}
         : { memory_at: parsed.memory_at }),
       ...(parsed.priority === undefined ? {} : { priority: parsed.priority }),
+      ...(userId === undefined ? {} : { user_id: userId }),
     });
     if (outcome.state !== "accepted") {
       emit(
@@ -957,9 +980,12 @@ async function runCommand(
                 state: "empty",
                 events: 0,
                 invalid_lines: report.invalid_lines,
+                unrecognized_lines: report.unrecognized_lines,
               })}\n`
           : renderLogText(report.events, report.invalid_lines),
       );
+      // Only lines that are not JSON are damage. Records an older build wrote
+      // are left out without a warning: after an upgrade they are expected.
       if (report.invalid_lines) {
         process.stderr.write(
           `${PRODUCT}: 跳过 ${report.invalid_lines} 行损坏的诊断记录。\n`,
@@ -1582,6 +1608,7 @@ async function runCommand(
           limit: parsed.limit,
           character_budget: parsed.characterBudget,
           timeout_ms: parsed.timeoutMs,
+          ...userIdParameters(userId),
         })
       : ({
           schema_version: 2,
@@ -1780,6 +1807,16 @@ async function runCommand(
   }
   if (named === "sources" && arguments_[1] !== "list") {
     const parsed = parseSourcesCommand(arguments_.slice(1));
+    if (
+      !parsed &&
+      arguments_.some((item) => item === "--user-id" || item === "--root")
+    ) {
+      return reject(
+        descriptor,
+        asJson,
+        "--user-id 和 --root 只能用于接入单个 hermes 来源（Hermes profile），用户 ID 只能是 1–64 位字母、数字、点、下划线或连字符。",
+      );
+    }
     if (!parsed) return usage();
     if (parsed.kind === "receipt") {
       const connected = await connectRuntime(environment);
@@ -2004,7 +2041,11 @@ async function runCommand(
       );
       return connected.exitCode;
     }
-    const outcome = await connected.reader<unknown>("/api/v2/user-profile");
+    const profileUser = userIdParameters(userId);
+    const outcome =
+      profileUser.user_id === undefined
+        ? await connected.reader<unknown>("/api/v2/user-profile")
+        : await connected.reader<unknown>("/api/v2/user-profile", profileUser);
     const value =
       outcome.state === "ok"
         ? (outcome.data as {
@@ -2109,7 +2150,7 @@ async function runCommand(
     }
     const { reader, operations } = connected;
     if (parsed.kind === "list") {
-      const document = await readMemoryPage(reader, parsed.query);
+      const document = await readMemoryPage(reader, parsed.query, userId);
       emit(
         asJson
           ? `${JSON.stringify(document)}\n`
@@ -2122,6 +2163,7 @@ async function runCommand(
         reader,
         parsed.memoryId,
         parsed.namespace,
+        userId,
       );
       emit(
         asJson
@@ -2133,8 +2175,8 @@ async function runCommand(
     if (parsed.kind === "add" || parsed.kind === "add-status") {
       const document =
         parsed.kind === "add"
-          ? await addMemory(operations, text ?? "", parsed.requestId)
-          : await readMemoryAddStatus(reader, parsed.requestId);
+          ? await addMemory(operations, text ?? "", parsed.requestId, userId)
+          : await readMemoryAddStatus(reader, parsed.requestId, userId);
       emit(
         asJson
           ? `${JSON.stringify(document)}\n`
@@ -2149,9 +2191,13 @@ async function runCommand(
             content: text,
             ...namespaceParameters(parsed.namespace),
             ...(parsed.task === undefined ? {} : { task: parsed.task }),
+            ...userIdParameters(userId),
           })
         : await operations.delete<unknown>(
-            withQuery(memoryPath, namespaceParameters(parsed.namespace)),
+            withQuery(memoryPath, {
+              ...namespaceParameters(parsed.namespace),
+              ...userIdParameters(userId),
+            }),
           );
     const document = {
       schema_version: 2,
@@ -3675,13 +3721,21 @@ async function connectSources(
       asJson,
     );
   }
-  if (parsed.importHistory) {
+  // Another user's Hermes profile is granted and imported in one connect
+  // request: only that request can carry a user and a root.
+  if (
+    parsed.importHistory ||
+    parsed.userId !== undefined ||
+    parsed.root !== undefined
+  ) {
     const document = await connectSourcesWithImport(reader, operations, {
       sourceIds: parsed.sourceIds,
       confirmOverrides: parsed.confirmOverrides,
       ...(parsed.requestId === undefined
         ? {}
         : { requestId: parsed.requestId }),
+      ...(parsed.userId === undefined ? {} : { userId: parsed.userId }),
+      ...(parsed.root === undefined ? {} : { root: parsed.root }),
     });
     emitFlowDocument(document, asJson);
     return flowExitCode(document);
@@ -5131,14 +5185,23 @@ type SourcesCommand =
       importHistory: boolean;
       confirmOverrides: boolean;
       requestId?: string;
+      /** `--user-id`: whose grant (Hermes only). Absent for `local`. */
+      userId?: string;
+      /** `--root`: an absolute root other than the candidate (Hermes only). */
+      root?: string;
     }
   | { kind: "all"; importHistory: boolean; confirmOverrides: boolean }
   | { kind: "receipt"; requestId: string };
 
 const SOURCE_ID_SHAPE = /^[a-z0-9][a-z0-9-]{0,127}$/;
 
+/** Only Hermes keeps one history per profile, so only Hermes may be
+ * connected for another user or from another root. */
+const MULTI_USER_SOURCES = new Set(["hermes"]);
+
 /** `sources connect <id> [<id> ...] | --all-detected [--import]
- * [--confirm-overrides] [--request-id <uuid>]` and `sources receipt <id>`.
+ * [--confirm-overrides] [--request-id <uuid>]
+ * [--user-id <id>] [--root <path>]` and `sources receipt <id>`.
  *
  * Sources may be named positionally or with repeated `--source` (the old
  * `init --source` spelling is rewritten to this command). */
@@ -5154,6 +5217,8 @@ function parseSourcesCommand(rest: readonly string[]): SourcesCommand | null {
   const sourceIds: string[] = [];
   const switches = new Set<string>();
   let requestId: string | undefined;
+  let userId: string | undefined;
+  let root: string | undefined;
   for (let index = 1; index < tokens.length; index += 1) {
     const token = tokens[index];
     if (["--import", "--confirm-overrides", "--all-detected"].includes(token)) {
@@ -5161,6 +5226,21 @@ function parseSourcesCommand(rest: readonly string[]): SourcesCommand | null {
       switches.add(token);
       continue;
     }
+    if (token === "--user-id" || token === "--root") {
+      const value = tokens[index + 1];
+      if (value === undefined || value.startsWith("--")) return null;
+      if (token === "--user-id") {
+        if (userId !== undefined || !isUserId(value)) return null;
+        userId = value;
+      } else {
+        if (root !== undefined) return null;
+        const expanded = expandRootPath(value);
+        if (expanded === null) return null;
+        root = expanded;
+      }
+      index += 1;
+      continue;
+    }
     if (token === "--source" || token === "--request-id") {
       const value = tokens[index + 1];
       if (value === undefined || value.startsWith("--")) return null;
@@ -5184,10 +5264,23 @@ function parseSourcesCommand(rest: readonly string[]): SourcesCommand | null {
   const confirmOverrides = switches.has("--confirm-overrides");
   if (
     requestId !== undefined &&
-    (!OPERATION_ID.test(requestId) || !importHistory)
+    (!OPERATION_ID.test(requestId) ||
+      !(importHistory || userId !== undefined || root !== undefined))
   ) {
     return null;
   }
+  // `--user-id local` is the default and changes nothing.
+  if (userId === DEFAULT_USER_ID) userId = undefined;
+  if (userId !== undefined || root !== undefined) {
+    // One Hermes profile per connect: the user and root name one grant.
+    if (
+      switches.has("--all-detected") ||
+      sourceIds.length !== 1 ||
+      !MULTI_USER_SOURCES.has(sourceIds[0])
+    ) {
+      return null;
+    }
+  }
   if (switches.has("--all-detected")) {
     // One names exact sources, the other means "whatever is detected".
     if (sourceIds.length > 0 || requestId !== undefined) return null;
@@ -5200,9 +5293,24 @@ function parseSourcesCommand(rest: readonly string[]): SourcesCommand | null {
     importHistory,
     confirmOverrides,
     ...(requestId === undefined ? {} : { requestId }),
+    ...(userId === undefined ? {} : { userId }),
+    ...(root === undefined ? {} : { root }),
   };
 }
 
+/** `--root` as an absolute path: a leading `~` is the home directory, a
+ * relative path is taken from the working directory. Core expands nothing. */
+function expandRootPath(value: string): string | null {
+  if (value.length === 0 || /[\0\r\n]/.test(value)) return null;
+  const expanded =
+    value === "~"
+      ? os.homedir()
+      : value.startsWith("~/")
+        ? path.join(os.homedir(), value.slice(2))
+        : value;
+  return path.resolve(expanded);
+}
+
 function parseGraphWidthCommand(
   rest: readonly string[],
 ): { decision: GraphWidthDecision; confirmed: boolean } | null {
@@ -6596,6 +6704,30 @@ const OWN_DATA_DIR_COMMANDS = new Set([
 
 /** Remove one `--data-dir <absolute path>` from argv; null when it is
  * repeated, missing its value, or not an absolute path. */
+/** Memory commands that honour the global `--user-id`. */
+const USER_ID_COMMANDS = new Set(["memories", "recall", "profile", "write"]);
+
+/**
+ * Take `--user-id <id>` out of the arguments. The default user (`local`) is
+ * returned as undefined, so a default request names no user at all and stays
+ * byte-identical to one sent before multi-user support.
+ */
+function liftUserIdOption(
+  arguments_: readonly string[],
+): { arguments_: string[]; userId?: string } | null {
+  const index = arguments_.indexOf("--user-id");
+  if (index === -1) return { arguments_: [...arguments_] };
+  if (arguments_.indexOf("--user-id", index + 1) !== -1) return null;
+  const value = arguments_[index + 1];
+  if (value === undefined || value.startsWith("--") || !isUserId(value)) {
+    return null;
+  }
+  return {
+    arguments_: [...arguments_.slice(0, index), ...arguments_.slice(index + 2)],
+    ...(value === DEFAULT_USER_ID ? {} : { userId: value }),
+  };
+}
+
 function liftDataDirOption(
   arguments_: readonly string[],
 ): { arguments_: string[]; dataDir?: string } | null {

memhub/apps/cli/src/command-catalog.ts

@@ -68,6 +68,12 @@ const JSON_OPTION: CommandOption = {
   flag: "--json",
   summary: "以 JSON 输出到 stdout，供脚本与 Agent 消费。",
 };
+/** Global to the memory commands: whose memories are read or written. */
+const USER_ID_OPTION: CommandOption = {
+  flag: "--user-id <id>",
+  summary:
+    "读写哪个用户的记忆，默认 local（App 用的就是 local）。1–64 位字母、数字、点、下划线或连字符；其他用户的记忆来自用 sources connect hermes --user-id 接入的 Hermes profile。",
+};
 const WATCH_OPTION: CommandOption = {
   flag: "--watch",
   summary: "串行持续刷新；Ctrl-C 只停止观察，不停止 Runtime。",
@@ -252,15 +258,25 @@ export const COMMANDS: readonly CommandDescriptor[] = [
   {
     name: "sources",
     summary:
-      "列出来源，直接接入来源，或按 request id 查接入回执。connect 不加 --import 只授权（处理保持冻结：不导入历史、不扫描新会话）；加 --import 同时导入历史并开启持续同步，返回 request id。目录来自环境变量覆盖的来源必须加 --confirm-overrides。",
+      "列出来源，直接接入来源，或按 request id 查接入回执。connect 不加 --import 只授权（处理保持冻结：不导入历史、不扫描新会话）；加 --import 同时导入历史并开启持续同步，返回 request id。目录来自环境变量覆盖的来源必须加 --confirm-overrides。Hermes 的每个 profile 是一个独立用户：用 --user-id 和 --root 接入，授权并导入历史（会自动排除 ~/.hermes/profiles/**，各用户互不读取）。",
     usage:
-      "memhub sources list [--json]\n            memhub sources connect <id> [<id> ...] [--import] [--confirm-overrides] [--request-id <uuid>] [--json]\n            memhub sources connect --all-detected [--import] [--confirm-overrides] [--json]\n            memhub sources receipt <request-id> [--json]",
+      "memhub sources list [--json]\n            memhub sources connect <id> [<id> ...] [--import] [--confirm-overrides] [--request-id <uuid>] [--json]\n            memhub sources connect hermes --user-id <id> --root <path> [--import] [--request-id <uuid>] [--json]\n            memhub sources connect --all-detected [--import] [--confirm-overrides] [--json]\n            memhub sources receipt <request-id> [--json]",
     options: [
       {
         flag: "--import",
         summary:
           "接入后立即导入历史会话并开启持续同步；不加时只授权，处理保持冻结。",
       },
+      {
+        flag: "--user-id <id>",
+        summary:
+          "只用于 hermes：这条接入属于哪个用户（默认 local）。和 --root 一起接入一个 Hermes profile，例如 --user-id wife --root ~/.hermes/profiles/wife；其他来源只能是 local。",
+      },
+      {
+        flag: "--root <path>",
+        summary:
+          "只用于 hermes：读取的根目录（~ 会展开为主目录）。默认是 Hermes 自己的目录 ~/.hermes。",
+      },
       {
         flag: "--all-detected",
         summary:
@@ -373,7 +389,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
     summary:
       "列出、搜索、查看、添加、修改或物理删除记忆（对话、工作、编码三个场景）。add 和 App 的「添加记忆」是同一个入口；delete 不可逆，需 --yes。",
     usage:
-      "memhub memories list [--scene <all|chat|working|coding>] [--source <id>] [--query <text>] [--status <state>] [--offset <n>] [--limit <n>] [--json]\n            memhub memories show <memory-id> [--kind <kind>] [--variant <variant>] [--json]\n            memhub memories add <text> [--request-id <uuid>] [--json]\n            memhub memories add --content-file <path> [--request-id <uuid>] [--json]\n            memhub memories add-status <request-id> [--json]\n            memhub memories update <memory-id> --content-file <path> [--kind <kind>] [--variant <variant>] [--task <text>] [--json]\n            memhub memories delete <memory-id> [--kind <kind>] [--variant <variant>] [--yes] [--json]",
+      "memhub memories list [--scene <all|chat|working|coding>] [--source <id>] [--query <text>] [--status <state>] [--offset <n>] [--limit <n>] [--user-id <id>] [--json]\n            memhub memories show <memory-id> [--kind <kind>] [--variant <variant>] [--user-id <id>] [--json]\n            memhub memories add <text> [--request-id <uuid>] [--user-id <id>] [--json]\n            memhub memories add --content-file <path> [--request-id <uuid>] [--user-id <id>] [--json]\n            memhub memories add-status <request-id> [--user-id <id>] [--json]\n            memhub memories update <memory-id> --content-file <path> [--kind <kind>] [--variant <variant>] [--task <text>] [--user-id <id>] [--json]\n            memhub memories delete <memory-id> [--kind <kind>] [--variant <variant>] [--yes] [--user-id <id>] [--json]",
     options: [
       {
         flag: "--scene <all|chat|working|coding>",
@@ -411,6 +427,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
           "add 的幂等键；用同一个 id 重试只会读回第一次的结果，不会重复添加。不给时自动生成并输出。",
       },
       DESTRUCTIVE_CONFIRM_OPTION,
+      USER_ID_OPTION,
       JSON_OPTION,
     ],
     exit_codes: RUNTIME_EXITS,
@@ -418,8 +435,8 @@ export const COMMANDS: readonly CommandDescriptor[] = [
   {
     name: "profile",
     summary: "读取 SDK 组装后的用户长期画像（L0 + 已启用的长期画像补充）。",
-    usage: "memhub profile [--json]",
-    options: [JSON_OPTION],
+    usage: "memhub profile [--user-id <id>] [--json]",
+    options: [USER_ID_OPTION, JSON_OPTION],
     exit_codes: RUNTIME_EXITS,
   },
   {
@@ -475,7 +492,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
     summary:
       "从原始对话与已沉淀记忆统一检索上下文；本地只读，不调用模型、不触发补索引。文本输出标明每条是「原始对话摘录」还是「记忆」（JSON 里是 kind: raw|memory）。",
     usage:
-      "memhub recall <query> [--kind <raw|memory>] [--source <id>] [--limit <n>] [--character-budget <n>] [--timeout-ms <n>] [--json]\n            memhub recall --query-file <path> [...]\n            memhub recall --stdin [...]",
+      "memhub recall <query> [--kind <raw|memory>] [--source <id>] [--limit <n>] [--character-budget <n>] [--timeout-ms <n>] [--user-id <id>] [--json]\n            memhub recall --query-file <path> [...]\n            memhub recall --stdin [...]",
     options: [
       {
         flag: "--query-file <path>",
@@ -502,6 +519,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
         flag: "--timeout-ms <n>",
         summary: "Core 查询时限 50..30000 毫秒，默认 2000。",
       },
+      USER_ID_OPTION,
       JSON_OPTION,
     ],
     exit_codes: RUNTIME_EXITS,
@@ -650,7 +668,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
     summary:
       "提交一段 OpenAI 格式的对话（messages）到写入队列，异步加工成记忆；或按 request id 查询状态。request id 是幂等键：不给时自动生成并输出，重试请带上同一个。新增一条结论式记忆请用 memories add。",
     usage:
-      "memhub write submit --messages-file <path> --session <id> [--source <id>] [--request-id <id>] [--memory-at <RFC3339>] [--live] [--json]\n            memhub write status <request-id> [--json]",
+      "memhub write submit --messages-file <path> --session <id> [--source <id>] [--request-id <id>] [--memory-at <RFC3339>] [--live] [--user-id <id>] [--json]\n            memhub write status <request-id> [--user-id <id>] [--json]",
     options: [
       {
         flag: "--messages-file <path>",
@@ -666,6 +684,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
       },
       { flag: "--memory-at <RFC3339>", summary: "对话发生时间，必须带时区。" },
       { flag: "--live", summary: "优先处理。" },
+      USER_ID_OPTION,
       JSON_OPTION,
     ],
     exit_codes: RUNTIME_EXITS,

memhub/apps/cli/src/context-search.ts

@@ -10,6 +10,8 @@ export interface ContextSearchInput {
   limit: number;
   character_budget: number;
   timeout_ms: number;
+  /** Whose memories to search. Omitted for the default user (`local`). */
+  user_id?: string;
 }
 
 /** The CLI's JSON document mirrors Core's SDK-native context schema v2. */

memhub/apps/cli/src/core-client.ts

@@ -12,6 +12,7 @@ const READ_ONLY_PATHS = [
   "/api/v1/live-sync",
   "/api/v1/memory-processing",
   "/api/v1/model-profiles",
+  "/api/v1/source-grants",
   "/api/v1/sources",
   "/api/v2/home-summary",
   "/api/v2/memory-library",
@@ -310,6 +311,8 @@ export async function submitMemoryWrite(
     messages: readonly OpenAIMessage[];
     memory_at?: string;
     priority?: "live" | "normal";
+    /** Whose memory this is. Omitted for the default user (`local`). */
+    user_id?: string;
   },
 ): Promise<MemoryWriteOutcome> {
   let response: { status: number; data: unknown };
@@ -327,6 +330,9 @@ export async function submitMemoryWrite(
         ...(request.priority === undefined
           ? {}
           : { priority: request.priority }),
+        ...(request.user_id === undefined || request.user_id === "local"
+          ? {}
+          : { user_id: request.user_id }),
       },
       {
         headers: { Authorization: `Bearer ${attachment.apiToken}` },
@@ -469,11 +475,16 @@ export async function deleteModelProfile(
 export async function readMemoryWrite(
   attachment: RuntimeAttachmentLike,
   requestId: string,
+  /** A receipt belongs to the user that submitted it; omit for `local`. */
+  userId?: string,
 ): Promise<MemoryWriteReceipt | null> {
   try {
     const response = await axios.get<MemoryWriteReceipt>(
       `${baseUrlOf(attachment)}/api/v2/memory-writes/${encodeURIComponent(requestId)}`,
       {
+        ...(userId === undefined || userId === "local"
+          ? {}
+          : { params: { user_id: userId } }),
         headers: { Authorization: `Bearer ${attachment.apiToken}` },
         timeout: 10_000,
         proxy: false,

memhub/apps/cli/src/core-operations.ts

@@ -6,9 +6,10 @@ const WRITE_PATHS = [
 ] as const;
 /** `POST /api/v2/memories` adds one memory (the app's 添加记忆). */
 const MEMORY_ADD_PATH = "/api/v2/memories";
-/** One SDK memory; delete names its namespace in the query string. */
+/** One SDK memory; delete names its namespace, and a non-default user, in
+ * the query string. */
 const SDK_MEMORY_PATH =
-  /^\/api\/v2\/memories\/[A-Za-z0-9][\w.-]{0,127}(?:\?kind=(?:chat|experience|fact)(?:&variant=(?:working|coding))?)?$/;
+  /^\/api\/v2\/memories\/[A-Za-z0-9][\w.-]{0,127}(?:\?(?:kind=(?:chat|experience|fact)(?:&variant=(?:working|coding))?(?:&user_id=[A-Za-z0-9._-]{1,64})?|user_id=[A-Za-z0-9._-]{1,64}))?$/;
 const MODEL_PROBE_PATH =
   /^\/api\/v1\/model-profiles\/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}\/probe$/;
 const MODEL_PROFILE_PATH =

memhub/apps/cli/src/headless-flows.ts

@@ -110,9 +110,18 @@ export async function connectSourcesWithImport(
     sourceIds: readonly string[];
     confirmOverrides: boolean;
     requestId?: string;
+    /** Whose grant this is. Omitted for the default user (`local`). */
+    userId?: string;
+    /** An absolute root other than the source's candidate (a Hermes
+     * profile). Core reads its identity itself, so the source's candidate
+     * root id and revision are not sent for it. */
+    root?: string;
   },
 ): Promise<FlowDocument> {
   const sourceIds = input.sourceIds;
+  // A second user's grant is not a first run, and a custom root has no
+  // candidate the app showed: both are a plain connect.
+  const profile = input.root !== undefined || input.userId !== undefined;
   if (
     sourceIds.length < 1 ||
     sourceIds.length > 32 ||
@@ -134,9 +143,14 @@ export async function connectSourcesWithImport(
     return invalid("source_unavailable");
   }
   const items = selected.map(([candidate]) => candidate);
-  const needsOverride = items
-    .filter((item) => item.requires_override_confirmation)
-    .map((item) => item.source_id);
+  // A custom root is the caller's own choice, not the environment override
+  // the confirmation exists for.
+  const needsOverride =
+    input.root !== undefined
+      ? []
+      : items
+          .filter((item) => item.requires_override_confirmation)
+          .map((item) => item.source_id);
   if (needsOverride.length > 0 && !input.confirmOverrides) {
     return {
       schema_version: 1,
@@ -154,18 +168,22 @@ export async function connectSourcesWithImport(
     request = buildSourceImportRequest({
       sourceIds: [...sourceIds],
       sourceRoots: Object.fromEntries(
-        items.map((item) => [item.source_id, item.candidate_root_id]),
+        items.flatMap((item) =>
+          input.root === undefined
+            ? [[item.source_id, item.candidate_root_id]]
+            : [],
+        ),
       ),
       sourceCandidateRevisions: Object.fromEntries(
         items.flatMap((item) =>
-          item.candidate_revision === undefined
+          item.candidate_revision === undefined || input.root !== undefined
             ? []
             : [[item.source_id, item.candidate_revision]],
         ),
       ),
       confirmEnvironmentOverrides: needsOverride.length > 0,
       fullRescan: false,
-      onboarding: true,
+      onboarding: !profile,
       enableSync: true,
       requestId,
     });
@@ -175,7 +193,11 @@ export async function connectSourcesWithImport(
     }
     throw error;
   }
-  const result = await operations.post<unknown>(request.path, request.body);
+  const result = await operations.post<unknown>(request.path, {
+    ...request.body,
+    ...(input.userId === undefined ? {} : { user_id: input.userId }),
+    ...(input.root === undefined ? {} : { root: input.root }),
+  });
   if (result.state === "rejected") {
     return rejected(requestId, result.code);
   }

memhub/apps/cli/src/log-report.ts

@@ -107,7 +107,11 @@ export async function readLogReport(
   query: LogQuery,
 ): Promise<{
   events: DiagnosticEventV1[];
+  /** Lines that are not JSON: the only ones `memhub logs` calls damaged. */
   invalid_lines: number;
+  /** Whole JSON lines the reader leaves out, such as an older build's
+   * records. Nothing is wrong with them, so the text output says nothing. */
+  unrecognized_lines: number;
   failure?: "log_directory_unreadable";
 }> {
   // `readDiagnosticEvents` treats an unreadable logs path as "no events". That
@@ -119,6 +123,7 @@ export async function readLogReport(
     return {
       events: [],
       invalid_lines: 0,
+      unrecognized_lines: 0,
       failure: "log_directory_unreadable",
     };
   }
@@ -129,6 +134,7 @@ export async function readLogReport(
     return {
       events: [],
       invalid_lines: 0,
+      unrecognized_lines: 0,
       failure: "log_directory_unreadable",
     };
   }

memhub/apps/cli/src/memory-report.ts

@@ -14,6 +14,25 @@ import { type CoreReader, RESERVED_MEMORY_SEGMENTS } from "./core-client.js";
 import type { CoreOperations } from "./core-operations.js";
 import { PRODUCT } from "./version.js";
 
+/** A MemHub user id, the same shape Core accepts. `local` is the default. */
+const USER_ID = /^[A-Za-z0-9._-]{1,64}$/;
+export const DEFAULT_USER_ID = "local";
+
+export function isUserId(value: string): boolean {
+  return USER_ID.test(value);
+}
+
+/**
+ * The `user_id` to send, or nothing. The default user is never named on the
+ * wire, so a request without `--user-id` stays exactly what it was before
+ * multi-user support and still works against an older Core.
+ */
+export function userIdParameters(userId?: string): { user_id?: string } {
+  return userId === undefined || userId === DEFAULT_USER_ID
+    ? {}
+    : { user_id: userId };
+}
+
 /** Output budget: a long memory is shortened visibly, never dropped silently. */
 const CONTENT_BUDGET = 240;
 
@@ -101,10 +120,12 @@ export interface MemoryDetailDocument {
 export async function readMemoryPage(
   reader: CoreReader,
   query: MemoryQuery,
+  userId?: string,
 ): Promise<MemoryPageDocument> {
   const parameters: Record<string, string | number> = {
     offset: query.offset,
     limit: query.limit,
+    ...userIdParameters(userId),
   };
   if (query.status !== undefined) parameters.statuses = query.status;
   if (query.source !== undefined) parameters.source = query.source;
@@ -156,8 +177,12 @@ export async function readMemoryDetail(
   reader: CoreReader,
   memoryId: string,
   namespace: MemoryNamespace = {},
+  userId?: string,
 ): Promise<MemoryDetailDocument> {
-  const parameters = namespaceParameters(namespace);
+  const parameters = {
+    ...namespaceParameters(namespace),
+    ...userIdParameters(userId),
+  };
   const outcome = await reader<unknown>(
     `/api/v2/memories/${memoryId}`,
     Object.keys(parameters).length > 0 ? parameters : undefined,
@@ -236,6 +261,7 @@ export async function addMemory(
   operations: CoreOperations,
   text: string,
   requestId: string = randomUUID(),
+  userId?: string,
 ): Promise<MemoryAddDocument> {
   const base = {
     schema_version: 2 as const,
@@ -259,7 +285,10 @@ export async function addMemory(
     }
     throw error;
   }
-  const result = await operations.post<unknown>(request.path, request.body);
+  const result = await operations.post<unknown>(request.path, {
+    ...request.body,
+    ...userIdParameters(userId),
+  });
   if (result.state !== "committed") {
     return { ...base, state: result.state, code: result.code };
   }
@@ -273,6 +302,7 @@ export async function addMemory(
 export async function readMemoryAddStatus(
   reader: CoreReader,
   requestId: string,
+  userId?: string,
 ): Promise<MemoryAddDocument> {
   const base = {
     schema_version: 2 as const,
@@ -283,7 +313,11 @@ export async function readMemoryAddStatus(
   if (!UUID.test(requestId)) {
     return { ...base, state: "invalid", code: "invalid_request_id" };
   }
-  const outcome = await reader<unknown>(memoryAddStatusPath(requestId));
+  const user = userIdParameters(userId);
+  const outcome =
+    user.user_id === undefined
+      ? await reader<unknown>(memoryAddStatusPath(requestId))
+      : await reader<unknown>(memoryAddStatusPath(requestId), user);
   if (outcome.state === "not_found") {
     return { ...base, state: "not_found", code: "memory_add_not_found" };
   }

memhub/apps/cli/src/source-progress-report.ts

@@ -38,6 +38,21 @@ export interface SourcesDocument {
   state: ReadState["state"];
   code?: string;
   sources: Array<Record<string, unknown>>;
+  /** Every grant of every user (`GET /api/v1/source-grants`). Absent when
+   * Core is too old to list them; `sources` is the default user's view. */
+  grants?: SourceGrant[];
+}
+
+/** One source grant. Only Hermes has grants for users other than `local`,
+ * one per Hermes profile root. */
+export interface SourceGrant {
+  user_id: string;
+  source_id: string;
+  root_label: string | null;
+  root_path: string | null;
+  authorized: boolean;
+  revoked_at: string | null;
+  sync_enabled: boolean;
 }
 
 export interface SyncDocument {
@@ -86,14 +101,66 @@ export async function readSources(
       sources: [],
     };
   }
+  const grants = await readSourceGrants(reader);
   return {
     schema_version: SCHEMA_VERSION,
     product: PRODUCT,
     state: "ok",
-    sources: result.data.map(projectSource),
+    // `/api/v1/sources` is the default user's view of every source.
+    sources: result.data.map((source) => ({
+      ...projectSource(source),
+      user_id: "local",
+    })),
+    ...(grants === null ? {} : { grants }),
   };
 }
 
+/** The grants of every user, or null when Core cannot list them (an older
+ * Core answers 404): the source list is then shown as before. */
+async function readSourceGrants(
+  reader: CoreReader,
+): Promise<SourceGrant[] | null> {
+  let outcome: ReadOutcome<unknown>;
+  try {
+    outcome = await reader<unknown>("/api/v1/source-grants");
+  } catch {
+    return null;
+  }
+  if (outcome.state !== "ok") return null;
+  const value = outcome.data;
+  if (!value || typeof value !== "object" || Array.isArray(value)) return null;
+  const items = (value as { grants?: unknown }).grants;
+  if (!Array.isArray(items)) return null;
+  const grants: SourceGrant[] = [];
+  const text = (candidate: unknown): string | null =>
+    typeof candidate === "string" && candidate.length <= 4096
+      ? candidate
+      : null;
+  for (const item of items) {
+    if (!item || typeof item !== "object" || Array.isArray(item)) return null;
+    const grant = item as Record<string, unknown>;
+    if (
+      typeof grant.user_id !== "string" ||
+      !/^[A-Za-z0-9._-]{1,64}$/.test(grant.user_id) ||
+      typeof grant.source_id !== "string" ||
+      !/^[a-z0-9][a-z0-9-]{0,127}$/.test(grant.source_id) ||
+      typeof grant.authorized !== "boolean"
+    ) {
+      return null;
+    }
+    grants.push({
+      user_id: grant.user_id,
+      source_id: grant.source_id,
+      root_label: text(grant.root_label),
+      root_path: text(grant.root_path),
+      authorized: grant.authorized,
+      revoked_at: text(grant.revoked_at),
+      sync_enabled: grant.sync_enabled === true,
+    });
+  }
+  return grants;
+}
+
 export async function readSync(reader: CoreReader): Promise<SyncDocument> {
   const result = validated(
     await reader<unknown>("/api/v1/live-sync"),
@@ -231,6 +298,22 @@ export function renderSourcesText(document: SourcesDocument): string {
     );
   }
   if (document.sources.length === 0) lines.push("（没有来源）");
+  // The list above is the default user's (`local`). Another user's grant,
+  // a Hermes profile, is listed with its user so it is not mistaken for it.
+  const others = (document.grants ?? []).filter(
+    (grant) => grant.user_id !== "local",
+  );
+  if (others.length > 0) {
+    lines.push("其他用户的连接：");
+    for (const grant of others) {
+      const name =
+        document.sources.find((source) => source.source_id === grant.source_id)
+          ?.display_name ?? grant.source_id;
+      lines.push(
+        `- ${name}（用户 ${grant.user_id}）：${grant.authorized ? "已连接" : grant.revoked_at ? "已撤销" : "未连接"}；处理${grant.sync_enabled ? "已开启" : "已冻结"}${grant.root_path ? `；目录 ${grant.root_path}` : ""}`,
+      );
+    }
+  }
   lines.push("");
   return lines.join("\n");
 }

memhub/apps/cli/src/source-progress.test.ts

@@ -179,12 +179,72 @@ describe("CLI-07/10 source and synchronization reads", () => {
       sync_enabled: true,
       sync_completeness: "additions_only",
     });
-    expect(requestedPaths()).toEqual(["/api/v1/sources"]);
+    // The grants list answers 404 here, as an older Core does: the source
+    // list is shown as before, every row the default user's.
+    expect(report.sources[0].user_id).toBe("local");
+    expect(report.grants).toBeUndefined();
+    expect(requestedPaths()).toEqual([
+      "/api/v1/sources",
+      "/api/v1/source-grants",
+    ]);
     expect(requestedPaths().some((path) => path.includes("detect"))).toBe(
       false,
     );
   });
 
+  it("shows every user's grant, a Hermes profile with its user", async () => {
+    const grants = {
+      schema_version: 1,
+      grants: [
+        {
+          user_id: "local",
+          source_id: "codex",
+          root_label: ".codex",
+          root_path: "/home/synthetic/.codex",
+          authorized: true,
+          revoked_at: null,
+          sync_enabled: true,
+        },
+        {
+          user_id: "wife",
+          source_id: "hermes",
+          root_label: "wife",
+          root_path: "/home/synthetic/.hermes/profiles/wife",
+          authorized: true,
+          revoked_at: null,
+          sync_enabled: true,
+        },
+      ],
+    };
+    transport.get.mockImplementation(async (url: string) =>
+      url === "/api/v1/source-grants"
+        ? { status: 200, data: grants }
+        : (RESPONSES.get(url) ?? { status: 404, data: {} }),
+    );
+    let streams = captureStreams();
+    try {
+      expect(await run(["sources", "list", "--json"], {})).toBe(0);
+    } finally {
+      streams.restore();
+    }
+    const report = JSON.parse(streams.stdout());
+    expect(report.grants).toEqual(grants.grants);
+    expect(report.sources[0]).toMatchObject({
+      source_id: "codex",
+      user_id: "local",
+    });
+    streams = captureStreams();
+    try {
+      expect(await run(["sources", "list"], {})).toBe(0);
+    } finally {
+      streams.restore();
+    }
+    expect(streams.stdout()).toContain("其他用户的连接：");
+    expect(streams.stdout()).toContain(
+      "hermes（用户 wife）：已连接；处理已开启；目录 /home/synthetic/.hermes/profiles/wife",
+    );
+  });
+
   it("keeps degraded synchronization distinct and prints its error code", async () => {
     const streams = captureStreams();
     try {
@@ -552,6 +612,7 @@ describe("CLI-19 progress and issue reads", () => {
       requestedPaths().every((path) =>
         [
           "/api/v1/sources",
+          "/api/v1/source-grants",
           "/api/v1/live-sync",
           "/api/v1/import-jobs",
           "/api/v1/memory-processing",

memhub/apps/cli/src/user-id-cli.test.ts

@@ -0,0 +1,405 @@
+import { mkdtemp, writeFile } from "node:fs/promises";
+import os, { tmpdir } from "node:os";
+import path from "node:path";
+import { beforeEach, describe, expect, it, vi } from "vitest";
+
+import { run } from "./cli.js";
+
+const transport = vi.hoisted(() => ({
+  get: vi.fn(),
+  post: vi.fn(),
+  put: vi.fn(),
+  patch: vi.fn(),
+  delete: vi.fn(),
+}));
+const direct = vi.hoisted(() => ({ get: vi.fn(), post: vi.fn() }));
+vi.mock("axios", () => ({
+  default: {
+    create: vi.fn(() => transport),
+    get: direct.get,
+    post: direct.post,
+    isAxiosError: () => false,
+  },
+}));
+
+const runtime = vi.hoisted(() => ({
+  inspectRuntimeStatus: vi.fn(),
+  discoverRuntimeAttachment: vi.fn(),
+}));
+vi.mock("@one-memory/runtime-control", async (importOriginal) => ({
+  ...(await importOriginal<typeof import("@one-memory/runtime-control")>()),
+  ...runtime,
+}));
+
+const REQUEST = "33333333-3333-4333-8333-333333333333";
+const HERMES = {
+  source_id: "hermes",
+  display_name: "Hermes",
+  storage_kinds: ["sqlite"],
+  available: true,
+  authorized: true,
+  detected: null,
+  root_label: ".hermes",
+  candidate_path: "~/.hermes",
+  candidate_kind: "default",
+  candidate_revision: 0,
+  candidate_root_id: "hermes-root-identity",
+  imported_messages: 0,
+  imported_conversations: 0,
+  capabilities: {
+    historical_import: "implemented",
+    incremental_import: "implemented",
+    live_sync: "implemented",
+    recall: "implemented",
+    installer: "not_implemented",
+    verification: "fixture_only",
+  },
+  sync_completeness: "additions_only",
+  location_state: "found",
+};
+const CODEX = {
+  ...HERMES,
+  source_id: "codex",
+  display_name: "Codex",
+  candidate_root_id: "codex-root-identity",
+};
+
+function addReceipt() {
+  return {
+    evidence_id: "44444444-4444-4444-8444-444444444444",
+    text: "Prefers tea",
+    scope: "all-apps",
+    created_at: "2026-10-27T00:00:00Z",
+    state: "pending",
+    task_id: null,
+    memory_ids: [],
+    error: null,
+  };
+}
+
+beforeEach(() => {
+  vi.clearAllMocks();
+  for (const method of [
+    ...Object.values(transport),
+    ...Object.values(direct),
+  ]) {
+    method.mockReset();
+  }
+  runtime.inspectRuntimeStatus.mockResolvedValue({
+    state: "running",
+    code: "runtime_verified",
+    instanceId: "instance-1",
+    host: "127.0.0.1",
+    port: 8765,
+    version: "1.2.3",
+  });
+  runtime.discoverRuntimeAttachment.mockResolvedValue({
+    instanceId: "instance-1",
+    apiToken: "synthetic-runtime-token",
+    host: "127.0.0.1",
+    port: 8765,
+    servicePid: 4242,
+    health: {
+      status: "ok",
+      service: "one-memory-core",
+      version: "1.2.3",
+      instance_id: "instance-1",
+    },
+  });
+  transport.get.mockImplementation(async (url: string) => {
+    if (url === "/api/v1/sources") {
+      return { status: 200, data: [HERMES, CODEX] };
+    }
+    if (url === "/api/v2/user-profile") {
+      return {
+        status: 200,
+        data: { request_id: "r-1", profile: null, elapsed_ms: 1 },
+      };
+    }
+    return { status: 404, data: {} };
+  });
+  transport.post.mockImplementation(
+    async (url: string, body: Record<string, unknown>) =>
+      url === "/api/v1/sources/connect"
+        ? {
+            status: 200,
+            data: {
+              request_id: body.request_id,
+              accepted_source_ids: ["hermes"],
+              failed_sources: {},
+              jobs: [],
+            },
+          }
+        : url === "/api/v2/memories"
+          ? { status: 201, data: addReceipt() }
+          : { status: 404, data: {} },
+  );
+  transport.put.mockResolvedValue({ status: 200, data: {} });
+  transport.delete.mockResolvedValue({ status: 204, data: "" });
+  direct.post.mockResolvedValue({
+    status: 202,
+    data: {
+      request_id: REQUEST,
+      state: "accepted",
+      created_at: "T1",
+      updated_at: "T1",
+    },
+  });
+  direct.get.mockResolvedValue({
+    data: {
+      request_id: REQUEST,
+      state: "accepted",
+      created_at: "T1",
+      updated_at: "T1",
+    },
+  });
+});
+
+describe("--user-id on the memory commands", () => {
+  it("passes the user as a query parameter to memory reads and deletes", async () => {
+    await capturedRun(["memories", "list", "--user-id", "wife", "--json"]);
+    expect(transport.get).toHaveBeenCalledWith("/api/v2/memory-library", {
+      params: { offset: 0, limit: 100, user_id: "wife" },
+    });
+    // Global: the flag may come before the subcommand too.
+    await capturedRun(["memories", "--user-id", "wife", "show", "m-1"]);
+    expect(transport.get).toHaveBeenCalledWith("/api/v2/memories/m-1", {
+      params: { user_id: "wife" },
+    });
+    await capturedRun(["memories", "add-status", REQUEST, "--user-id", "wife"]);
+    expect(transport.get).toHaveBeenCalledWith(
+      `/api/v2/memories/additions/${REQUEST}`,
+      { params: { user_id: "wife" } },
+    );
+    await capturedRun(["profile", "--user-id", "wife", "--json"]);
+    expect(transport.get).toHaveBeenCalledWith("/api/v2/user-profile", {
+      params: { user_id: "wife" },
+    });
+    await capturedRun([
+      "memories",
+      "delete",
+      "fact-1",
+      "--kind",
+      "fact",
+      "--yes",
+      "--user-id",
+      "wife",
+    ]);
+    expect(transport.delete).toHaveBeenCalledWith(
+      "/api/v2/memories/fact-1?kind=fact&user_id=wife",
+      undefined,
+    );
+    await capturedRun([
+      "memories",
+      "delete",
+      "m-1",
+      "--yes",
+      "--user-id",
+      "wife",
+    ]);
+    expect(transport.delete).toHaveBeenCalledWith(
+      "/api/v2/memories/m-1?user_id=wife",
+      undefined,
+    );
+    await capturedRun(["write", "status", REQUEST, "--user-id", "wife"]);
+    expect(direct.get.mock.calls[0][0]).toBe(
+      `http://127.0.0.1:8765/api/v2/memory-writes/${REQUEST}`,
+    );
+    expect(direct.get.mock.calls[0][1].params).toEqual({ user_id: "wife" });
+  });
+
+  it("passes the user in the body of memory writes and searches", async () => {
+    await capturedRun(["memories", "add", "Prefers tea", "--user-id", "wife"]);
+    expect(transport.post.mock.calls[0][0]).toBe("/api/v2/memories");
+    expect(transport.post.mock.calls[0][1]).toMatchObject({
+      text: "Prefers tea",
+      scope: "all-apps",
+      user_id: "wife",
+    });
+    const directory = await mkdtemp(path.join(tmpdir(), "om-user-id-"));
+    const content = path.join(directory, "content.txt");
+    await writeFile(content, "Prefers green tea\n");
+    await capturedRun([
+      "memories",
+      "update",
+      "m-1",
+      "--content-file",
+      content,
+      "--user-id",
+      "wife",
+    ]);
+    expect(transport.put.mock.calls[0][0]).toBe("/api/v2/memories/m-1");
+    expect(transport.put.mock.calls[0][1]).toMatchObject({ user_id: "wife" });
+    await capturedRun(["recall", "tea", "--user-id", "wife", "--json"]);
+    const search = transport.post.mock.calls.find(
+      ([url]) => url === "/api/v2/context/search",
+    );
+    expect(search?.[1]).toMatchObject({ query: "tea", user_id: "wife" });
+    const messages = path.join(directory, "messages.json");
+    await writeFile(
+      messages,
+      JSON.stringify([{ role: "user", content: "I like tea" }]),
+    );
+    await capturedRun([
+      "write",
+      "submit",
+      "--messages-file",
+      messages,
+      "--session",
+      "s-1",
+      "--user-id",
+      "wife",
+    ]);
+    expect(direct.post.mock.calls[0][0]).toBe(
+      "http://127.0.0.1:8765/api/v2/memory-writes",
+    );
+    expect(direct.post.mock.calls[0][1]).toMatchObject({
+      session_id: "s-1",
+      user_id: "wife",
+    });
+  });
+
+  it("names no user by default or for --user-id local", async () => {
+    for (const extra of [[], ["--user-id", "local"]]) {
+      transport.get.mockClear();
+      transport.post.mockClear();
+      await capturedRun(["memories", "list", ...extra, "--json"]);
+      expect(transport.get).toHaveBeenCalledWith("/api/v2/memory-library", {
+        params: { offset: 0, limit: 100 },
+      });
+      await capturedRun(["profile", ...extra]);
+      expect(transport.get).toHaveBeenCalledWith(
+        "/api/v2/user-profile",
+        undefined,
+      );
+      await capturedRun(["memories", "add", "Prefers tea", ...extra]);
+      expect(transport.post.mock.calls[0][1]).not.toHaveProperty("user_id");
+      await capturedRun(["recall", "tea", ...extra, "--json"]);
+      const search = transport.post.mock.calls.find(
+        ([url]) => url === "/api/v2/context/search",
+      );
+      expect(search?.[1]).not.toHaveProperty("user_id");
+    }
+  });
+
+  it("refuses a malformed user id before contacting Core", async () => {
+    for (const value of ["bad:id", "", "x".repeat(65), "a b", "--json"]) {
+      const result = await capturedRun([
+        "memories",
+        "list",
+        "--user-id",
+        value,
+      ]);
+      expect(result.code).toBe(2);
+    }
+    expect(
+      (await capturedRun(["profile", "--user-id", "a", "--user-id", "b"])).code,
+    ).toBe(2);
+    expect((await capturedRun(["recall", "tea", "--user-id"])).code).toBe(2);
+    expect(transport.get).not.toHaveBeenCalled();
+    expect(transport.post).not.toHaveBeenCalled();
+  });
+});
+
+describe("sources connect for a Hermes profile", () => {
+  it("connects hermes for another user from its own root", async () => {
+    const result = await capturedRun([
+      "sources",
+      "connect",
+      "hermes",
+      "--user-id",
+      "wife",
+      "--root",
+      "~/.hermes/profiles/wife",
+      "--json",
+    ]);
+    expect(result.code).toBe(0);
+    expect(transport.post).toHaveBeenCalledTimes(1);
+    const [url, body] = transport.post.mock.calls[0];
+    expect(url).toBe("/api/v1/sources/connect");
+    expect(body).toMatchObject({
+      source_ids: ["hermes"],
+      // Core reads the custom root's identity itself.
+      source_roots: {},
+      source_candidate_revisions: {},
+      onboarding: false,
+      user_id: "wife",
+      root: path.join(os.homedir(), ".hermes/profiles/wife"),
+    });
+  });
+
+  it("keeps a plain connect unchanged and accepts --import with a profile", async () => {
+    await capturedRun(["sources", "connect", "hermes", "--import", "--json"]);
+    const plain = transport.post.mock.calls[0][1];
+    expect(plain).not.toHaveProperty("user_id");
+    expect(plain).not.toHaveProperty("root");
+    expect(plain).toMatchObject({
+      source_roots: { hermes: "hermes-root-identity" },
+      onboarding: true,
+    });
+    transport.post.mockClear();
+    const result = await capturedRun([
+      "sources",
+      "connect",
+      "hermes",
+      "--import",
+      "--user-id",
+      "wife",
+      "--root",
+      "/profiles/wife",
+      "--request-id",
+      REQUEST,
+      "--json",
+    ]);
+    expect(result.code).toBe(0);
+    expect(transport.post.mock.calls[0][1]).toMatchObject({
+      request_id: REQUEST,
+      user_id: "wife",
+      root: "/profiles/wife",
+    });
+  });
+
+  it("refuses --user-id and --root for any source but hermes", async () => {
+    for (const argv of [
+      ["sources", "connect", "codex", "--user-id", "wife"],
+      ["sources", "connect", "codex", "--root", "/somewhere", "--import"],
+      ["sources", "connect", "hermes", "codex", "--user-id", "wife"],
+      ["sources", "connect", "--all-detected", "--user-id", "wife"],
+      ["sources", "connect", "hermes", "--user-id", "bad:id"],
+    ]) {
+      const result = await capturedRun(argv);
+      expect(result.code).toBe(2);
+      expect(result.stderr).toContain("hermes");
+    }
+    expect(transport.post).not.toHaveBeenCalled();
+  });
+});
+
+async function capturedRun(arguments_: string[]) {
+  let stdout = "";
+  let stderr = "";
+  const stdoutSpy = vi
+    .spyOn(process.stdout, "write")
+    .mockImplementation((chunk) => {
+      stdout += String(chunk);
+      return true;
+    });
+  const stderrSpy = vi
+    .spyOn(process.stderr, "write")
+    .mockImplementation((chunk) => {
+      stderr += String(chunk);
+      return true;
+    });
+  try {
+    return {
+      code: await run(arguments_, {
+        ONE_MEMORY_DATA_DIR: "/synthetic/data",
+      }),
+      stdout,
+      stderr,
+    };
+  } finally {
+    stdoutSpy.mockRestore();
+    stderrSpy.mockRestore();
+  }
+}

memhub/apps/core/pyproject.toml

@@ -12,7 +12,7 @@ dependencies = [
     "fastapi>=0.141.1",
     "filelock>=3.32.4",
     # Public SDK version validated against MemHub's permission and recall contracts.
-    "hy-memory==2.0.8",
+    "hy-memory==2.0.9",
     # chromadb arrives through the HY SDK; locked here so an unrelated upgrade
     # cannot move the SDK's store. Built-in embedding consumes numpy directly.
     "chromadb==1.5.9",

memhub/apps/core/src/one_memory/api.py

@@ -14,6 +14,7 @@ from collections.abc import (
     MutableMapping,
 )
 from contextlib import asynccontextmanager, nullcontext, suppress
+from dataclasses import replace
 from datetime import UTC, datetime, timedelta
 from functools import partial
 from pathlib import Path
@@ -117,6 +118,7 @@ from one_memory.imports.access import (
     AuthorizedSourceRoot,
     SourceAccessError,
     root_identity_matches,
+    source_root_identity,
 )
 from one_memory.imports.adapters import (
     ScanFailure,
@@ -157,6 +159,7 @@ from one_memory.imports.sources.common import SourceInstallState
 from one_memory.imports.sources.registry import (
     RootCandidateKind,
     SourceRegistration,
+    SourceRootCandidate,
     build_source_catalog,
     workbuddy_default_homes,
 )
@@ -244,7 +247,17 @@ from one_memory.service_levels import (
 from one_memory.source_import_receipts import SourceImportReceipt, build_source_import_receipt
 from one_memory.source_read_access import SourceReadAccess
 from one_memory.source_totals import SourceTotals
+from one_memory.strict_query import forbid_unknown_query_params
 from one_memory.target_schema import bootstrap_fresh_management_schema
+from one_memory.user_scope import (
+    DEFAULT_USER_ID,
+    MULTI_USER_SOURCES,
+    USER_ID_PATTERN,
+    UserId,
+    lane_key,
+    owns,
+    split_lane,
+)
 from one_memory.workspace_resolution import resolve_workspace_scopes
 
 logger = logging.getLogger(__name__)
@@ -387,6 +400,12 @@ class CreateImportJobRequest(BaseModel):
 
 class ConnectSourcesRequest(BaseModel):
     source_ids: tuple[str, ...] = Field(min_length=1, max_length=32)
+    #: Whose grant this is. Only Hermes may connect a root for a user other
+    #: than `local`: a Hermes profile is another person's history.
+    user_id: UserId = DEFAULT_USER_ID
+    #: A root chosen by the caller instead of the source's candidate, such as
+    #: `~/.hermes/profiles/<name>`. Hermes only; Core computes its identity.
+    root: str | None = Field(default=None, min_length=1, max_length=4096)
     confirm_environment_overrides: bool = False
     full_rescan: bool = False
     onboarding: bool = False
@@ -422,6 +441,8 @@ class SdkMemoryUpdateRequest(BaseModel):
     variant: SdkMemoryVariant | None = None
     # Experiences only: also replace the task.
     task: str | None = Field(default=None, min_length=1, max_length=200_000)
+    #: Whose memory this is; another user's id answers 404.
+    user_id: UserId = DEFAULT_USER_ID
 
 
 class SdkUserProfileResponse(BaseModel):
@@ -584,11 +605,15 @@ class ManualExtractReceipt(BaseModel):
 
 
 class AskMemoryRequest(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
     question: str = Field(min_length=1, max_length=2_000)
     scope: str = Field(default="all-apps", min_length=1, max_length=512)
     # Host-injected project directory, same boundary as recall. Optional so
     # the web UI and older callers keep working unchanged.
     workspace_path: str | None = Field(default=None, max_length=4_096)
+    #: Whose memories answer. Omitted, it is `local`, the App's library.
+    user_id: UserId = DEFAULT_USER_ID
 
 
 class SdkHomeSummary(BaseModel):
@@ -626,6 +651,27 @@ class MemoryGraphResponse(BaseModel):
     hy_graph_available: bool = True
 
 
+class SourceGrantView(BaseModel):
+    """One `(user_id, source_id)` grant, as `GET /api/v1/source-grants` lists it."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    user_id: str
+    source_id: str
+    root_label: str
+    root_path: str
+    authorized: bool
+    revoked_at: datetime | None
+    sync_enabled: bool
+
+
+class SourceGrantList(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    schema_version: Literal[1] = 1
+    grants: tuple[SourceGrantView, ...]
+
+
 class SdkMigrationRequired(RuntimeError):
     """A serving Core was started on a home that is not on SDK authority."""
 
@@ -928,6 +974,7 @@ def create_app(
                 write_queue=memory_write_queue,
                 extraction_settings=app.state.extraction_settings_store,
                 data_dir=runtime_settings.data_dir,
+                user_ids=lambda: _known_user_ids(app),
             )
             # C2 owns the single production budget (handle 5s + reserve 250ms,
             # total 5.25s, lease 10s, renew 1s, claim 250ms). The composition
@@ -1100,6 +1147,13 @@ def create_app(
                 source_id
                 for source_id, registration in app.state.source_catalog.items()
                 if registration.capabilities.live_sync is CapabilityState.IMPLEMENTED
+            ) + tuple(
+                lane
+                for lane in _other_user_lanes(
+                    app.state.source_catalog, app.state.source_grant_store
+                )
+                if app.state.source_catalog[split_lane(lane)[1]].capabilities.live_sync
+                is CapabilityState.IMPLEMENTED
             )
             # Resolved before the supervisor is built so it can state truthfully
             # how it is being driven: watching only applies to roots that exist
@@ -1496,9 +1550,12 @@ def create_app(
         Depends(local_api_guard(runtime_settings, valid_ui_session=ui_session_valid)),
         Depends(require_external_endpoint),
     ]
-    app.include_router(memory_add_router(protected))
+    # Memory reads and writes refuse a query parameter they do not declare: an
+    # ignored `user_id` would read as isolation that never happened.
+    strict = [*protected, Depends(forbid_unknown_query_params)]
+    app.include_router(memory_add_router(strict))
     app.include_router(file_import_router(protected))
-    app.include_router(sdk_context_router(protected))
+    app.include_router(sdk_context_router(strict))
 
     # A4-01: scan-root configuration, so the CLI's `path` command has something
     # to read instead of refusing.
@@ -1512,7 +1569,7 @@ def create_app(
 
     if getattr(app.state, "memory_write_registry", None) is None:
         app.state.memory_write_registry = WriteRegistry()
-    app.include_router(sdk_memory_write_router(protected))
+    app.include_router(sdk_memory_write_router(strict))
     from one_memory.call_event_routes import call_event_router
 
     app.include_router(call_event_router(protected))
@@ -2287,7 +2344,11 @@ def create_app(
         return SourceSyncState(source_id=source_id, sync_enabled=request.enabled)
 
     async def _prepare_source_resume(source_id: str) -> None:
-        """Prepare catch-up work before committing the new desired state."""
+        """Prepare catch-up work before committing the new desired state.
+
+        `source_id` is a lane key, so another user's lane prepares that
+        user's own cursors and history job.
+        """
         await _live_sync(app).initialize_source(source_id, baseline_missing_files=True)
         active = {
             ImportStatus.PENDING,
@@ -2297,14 +2358,17 @@ def create_app(
             ImportStatus.FAILED,
         }
         if any(
-            job.source_id == source_id and job.operation == "history" and job.status in active
+            lane_key(job.user_id, job.source_id) == source_id
+            and job.operation == "history"
+            and job.status in active
             for job in _job_store(app).list_jobs()
         ):
             return
+        user_id, bare_source_id = split_lane(source_id)
         await asyncio.to_thread(
             _job_store(app).create,
-            user_id="local",
-            source_id=source_id,
+            user_id=user_id,
+            source_id=bare_source_id,
             operation="history",
         )
 
@@ -2312,7 +2376,7 @@ def create_app(
         """Run work prepared for a source after config.json has committed."""
         try:
             for job in _job_store(app).list_jobs():
-                if job.source_id != source_id:
+                if lane_key(job.user_id, job.source_id) != source_id:
                     continue
                 if job.status is ImportStatus.PENDING:
                     _supervisor(app).enqueue(job.job_id)
@@ -2341,8 +2405,13 @@ def create_app(
         response_model=SourceGrantReceipt,
         dependencies=protected,
     )
-    async def revoke_source(source_id: str) -> SourceGrantReceipt:
+    async def revoke_source(
+        source_id: str,
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
+    ) -> SourceGrantReceipt:
         _require_registration(app, source_id)
+        if user_id != DEFAULT_USER_ID:
+            return await _revoke_other_user_grant(source_id, user_id)
         receipt = _grant_store(app).receipt(
             user_id="local",
             source_id=source_id,
@@ -2378,6 +2447,65 @@ def create_app(
         await _refresh_live_watch(app)
         return receipt
 
+    async def _revoke_other_user_grant(source_id: str, user_id: str) -> SourceGrantReceipt:
+        """Stop reading another user's root. The App's switch is left alone."""
+        if source_id not in MULTI_USER_SOURCES:
+            raise HTTPException(status_code=422, detail="user_id_requires_multi_user_source")
+        lane = lane_key(user_id, source_id)
+        receipt = await asyncio.to_thread(
+            partial(_grant_store(app).receipt, user_id=user_id, source_id=source_id)
+        )
+        if receipt is None:
+            raise HTTPException(status_code=404, detail="grant_not_found")
+        active_statuses = {
+            ImportStatus.PENDING,
+            ImportStatus.RUNNING,
+            ImportStatus.PAUSED,
+            ImportStatus.INTERRUPTED,
+            ImportStatus.FAILED,
+        }
+        for job in _job_store(app).list_jobs():
+            if lane_key(job.user_id, job.source_id) == lane and job.status in active_statuses:
+                await _supervisor(app).cancel(job.job_id)
+        await _live_sync(app).stop_source(lane)
+        _supervisor(app).unregister_adapter(lane)
+        receipt = await asyncio.to_thread(
+            partial(_grant_store(app).revoke, user_id=user_id, source_id=source_id)
+        )
+        await _refresh_live_watch(app)
+        return receipt
+
+    @app.get(
+        "/api/v1/source-grants",
+        response_model=SourceGrantList,
+        dependencies=protected,
+    )
+    async def list_source_grants() -> SourceGrantList:
+        """Every grant of every user. `/api/v1/sources` stays `local`'s, as the App reads it."""
+
+        def describe() -> SourceGrantList:
+            config_store = _config_store(app)
+            views = []
+            for receipt, root_path in _grant_store(app).grants():
+                lane = lane_key(receipt.user_id, receipt.source_id)
+                authorized = receipt.revoked_at is None and receipt.revision > 0
+                views.append(
+                    SourceGrantView(
+                        user_id=receipt.user_id,
+                        source_id=receipt.source_id,
+                        root_label=receipt.root_label,
+                        root_path=_display_candidate_path(
+                            Path(root_path), runtime_settings.source_home
+                        ),
+                        authorized=authorized,
+                        revoked_at=receipt.revoked_at,
+                        sync_enabled=authorized and _lane_sync_enabled(config_store, lane),
+                    )
+                )
+            return SourceGrantList(grants=tuple(views))
+
+        return await asyncio.to_thread(describe)
+
     @app.post(
         "/api/v1/import-jobs",
         response_model=ImportJob,
@@ -2406,6 +2534,8 @@ def create_app(
     async def connect_sources(
         request: ConnectSourcesRequest,
     ) -> ConnectSourcesResponse:
+        if request.user_id != DEFAULT_USER_ID or request.root is not None:
+            return await _connect_other_root(request)
         if request.onboarding:
             app.state.history_import_gate_released = False
         source_ids = tuple(dict.fromkeys(request.source_ids))
@@ -2511,6 +2641,103 @@ def create_app(
                 raise
         return ConnectSourcesResponse(**result.model_dump())
 
+    async def _connect_other_root(request: ConnectSourcesRequest) -> ConnectSourcesResponse:
+        """Grant one Hermes root to a user: `~/.hermes/profiles/<name>` for its own user.
+
+        Only Hermes holds more than one person's history (one profile per
+        person), so a non-`local` user or a caller-chosen root is refused for
+        every other source. The grant is `(user_id, source_id)`; its cursors,
+        jobs and live sync follow that key.
+        """
+        source_ids = tuple(dict.fromkeys(request.source_ids))
+        if any(source_id not in MULTI_USER_SOURCES for source_id in source_ids):
+            raise HTTPException(status_code=422, detail="user_id_requires_multi_user_source")
+        if request.root is not None and len(source_ids) != 1:
+            raise HTTPException(status_code=422, detail="source_root_requires_one_source")
+        if request.user_id != DEFAULT_USER_ID and request.root is None:
+            # The user is the root that is read: another user must name theirs,
+            # or they would be handed `local`'s history.
+            raise HTTPException(status_code=422, detail="source_root_required_for_user")
+        if request.onboarding or request.import_upload_id is not None:
+            raise HTTPException(status_code=422, detail="user_id_unsupported_for_request")
+        if request.request_id is None:
+            raise HTTPException(status_code=409, detail="source_confirmation_required")
+        registrations = []
+        source_roots = dict(request.source_roots)
+        for source_id in source_ids:
+            registration = _require_registration(app, source_id)
+            if request.root is not None:
+                root = Path(request.root)
+                if not root.is_absolute():
+                    raise HTTPException(status_code=422, detail="source_root_must_be_absolute")
+                # Named outright by the caller, like an environment override:
+                # no saved candidate may replace it, and naming it confirms it.
+                registration = replace(
+                    registration,
+                    candidate=SourceRootCandidate(
+                        path=root, kind=RootCandidateKind.ENVIRONMENT_OVERRIDE
+                    ),
+                )
+                try:
+                    source_roots[source_id] = await asyncio.to_thread(source_root_identity, root)
+                except SourceAccessError as error:
+                    raise HTTPException(status_code=409, detail=error.code) from None
+            registrations.append(registration)
+        existing_intent = await asyncio.to_thread(
+            partial(
+                app.state.source_connection_store.get,
+                request.request_id,
+                user_id=request.user_id,
+            )
+        )
+        try:
+            result = await asyncio.to_thread(
+                partial(
+                    app.state.source_connection_store.connect,
+                    request_id=request.request_id,
+                    registrations=tuple(registrations),
+                    source_roots=source_roots,
+                    confirm_overrides=(
+                        request.confirm_environment_overrides or request.root is not None
+                    ),
+                    full_rescan=request.full_rescan,
+                    user_id=request.user_id,
+                )
+            )
+        except ConnectionIntentError as error:
+            raise HTTPException(status_code=409, detail=str(error)) from None
+        if existing_intent is not None:
+            return ConnectSourcesResponse(**result.model_dump())
+        for job in result.jobs:
+            registration = _require_registration(app, job.source_id)
+            grant = await asyncio.to_thread(
+                partial(
+                    _grant_store(app).authorized_root,
+                    user_id=job.user_id,
+                    source_id=job.source_id,
+                )
+            )
+            if grant is None:
+                continue
+            lane = lane_key(job.user_id, job.source_id)
+            _supervisor(app).register_adapter(
+                registration.create_adapter(grant), user_id=job.user_id
+            )
+            _detection(app).pop(lane, None)
+            if job.user_id != DEFAULT_USER_ID:
+                if registration.capabilities.live_sync is CapabilityState.IMPLEMENTED:
+                    _live_sync(app).ensure_source(lane)
+                # Another user's grant syncs while it stands (see
+                # `_lane_sync_enabled`); the App's switch is not touched.
+                await _prepare_source_resume(lane)
+                await _activate_source_resume(lane)
+            elif request.enable_sync:
+                await _prepare_source_resume(lane)
+                await asyncio.to_thread(_config_store(app).set_sync, job.source_id, True)
+                await _activate_source_resume(lane)
+        await _refresh_live_watch(app)
+        return ConnectSourcesResponse(**result.model_dump())
+
     @app.get(
         "/api/v1/source-connections/{request_id}",
         response_model=ConnectionIntentReceipt,
@@ -2527,8 +2754,12 @@ def create_app(
         response_model=list[ImportJob],
         dependencies=protected,
     )
-    def list_import_jobs() -> list[ImportJob]:
-        return _job_store(app).list_jobs()
+    def list_import_jobs(
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
+    ) -> list[ImportJob]:
+        # One user's jobs: the App reads `local`'s and must not show another
+        # user's Hermes import as its own.
+        return [job for job in _job_store(app).list_jobs() if job.user_id == user_id]
 
     @app.get(
         "/api/v1/live-sync",
@@ -2857,9 +3088,11 @@ def create_app(
     @app.get(
         "/api/v2/home-summary",
         response_model=SdkHomeSummary,
-        dependencies=protected,
+        dependencies=strict,
     )
-    async def sdk_home_summary() -> SdkHomeSummary:
+    async def sdk_home_summary(
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
+    ) -> SdkHomeSummary:
         """The home card: the memory page's count and its newest memory.
 
         The count is `memory_stats` total (chat + working + coding
@@ -2871,12 +3104,12 @@ def create_app(
             client = await _sdk_client_handle(app).client()
             stats = await asyncio.to_thread(
                 client.memory_stats,
-                user_id="local",
+                user_id=user_id,
                 statuses=["active"],
             )
             raw = await asyncio.to_thread(
                 client.list_memories,
-                user_id="local",
+                user_id=user_id,
                 scene="all",
                 statuses=["active"],
                 limit=1,
@@ -2897,7 +3130,7 @@ def create_app(
     @app.get(
         "/api/v2/memory-library",
         response_model=HySdkMemoryPage,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def sdk_memory_library(
         statuses: list[str] = Query(default=["active"]),  # noqa: B008
@@ -2907,6 +3140,7 @@ def create_app(
         limit: int = Query(default=LIBRARY_PAGE_SIZE, ge=1, le=500),
         window_id: str | None = Query(default=None, min_length=1, max_length=64),
         scene: Literal["all", "chat", "working", "coding"] = Query(default="all"),
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
     ) -> HySdkMemoryPage:
         """One page of memories across chat, working and coding.
 
@@ -2941,10 +3175,11 @@ def create_app(
                     limit=limit,
                     window_id=window_id,
                     scene=scene,
+                    user_id=user_id,
                 )
             raw = await asyncio.to_thread(
                 client.list_memories,
-                user_id="local",
+                user_id=user_id,
                 scene=scene,
                 statuses=list(statuses),
                 include_sources=[source] if source else None,
@@ -2965,12 +3200,13 @@ def create_app(
     @app.get(
         "/api/v2/memories/graph",
         response_model=MemoryGraphResponse,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def sdk_memory_graph(
         limit: int = Query(default=160, ge=1, le=300),
         offset: int = Query(default=0, ge=0),
         source: str | None = Query(default=None, min_length=1),
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
     ) -> MemoryGraphResponse:
         """The memory map on SDK authority: SDK memories plus the Kuzu overlay.
 
@@ -2984,7 +3220,7 @@ def create_app(
             # left list's 20-row pagination window.
             raw = await asyncio.to_thread(
                 client.list_memories,
-                user_id="local",
+                user_id=user_id,
                 scene="all",
                 statuses=["active"],
                 include_sources=[source] if source else None,
@@ -3054,7 +3290,9 @@ def create_app(
         # source withholds the overlay, as on the legacy map.
         if not await asyncio.to_thread(_read_access(app).has_revoked_sources):
             try:
-                hy_graph = await _sdk_memory_runtime(app).graph_snapshot(limit=limit)
+                hy_graph = await _sdk_memory_runtime(app).graph_snapshot(
+                    limit=limit, user_id=user_id
+                )
             except HySdkError:
                 hy_graph = None
         if hy_graph is not None:
@@ -3104,12 +3342,14 @@ def create_app(
     @app.get(
         "/api/v2/user-profile",
         response_model=SdkUserProfileResponse,
-        dependencies=protected,
+        dependencies=strict,
     )
-    async def sdk_user_profile() -> SdkUserProfileResponse:
+    async def sdk_user_profile(
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
+    ) -> SdkUserProfileResponse:
         try:
             client = await _sdk_client_handle(app).client()
-            value = await asyncio.to_thread(client.get_user_profile, user_id="local")
+            value = await asyncio.to_thread(client.get_user_profile, user_id=user_id)
             profile = HySdkUserProfile.from_sdk(value)
         except Exception as error:
             raise _sdk_http_error(error, "hy_memory_profile_read_failed") from None
@@ -3118,18 +3358,23 @@ def create_app(
     @app.get(
         "/api/v2/memories/{memory_id}",
         response_model=HySdkMemoryRecord,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def get_sdk_memory(
         memory_id: str,
         kind: SdkMemoryKind | None = None,
         variant: SdkMemoryVariant | None = None,
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
     ) -> HySdkMemoryRecord:
         _check_memory_namespace(kind, variant)
         try:
             memory = await _get_sdk_memory(
                 _sdk_client_handle(app), memory_id, kind=kind, variant=variant
             )
+            # The SDK's `get` takes no user: another user's memory reads as
+            # missing, never as existing.
+            if memory is not None and not owns(memory.user_id, user_id):
+                memory = None
         except AmbiguousMemoryIdError:
             raise HTTPException(status_code=409, detail="memory_id_ambiguous") from None
         except Exception as error:
@@ -3141,7 +3386,7 @@ def create_app(
     @app.put(
         "/api/v2/memories/{memory_id}",
         response_model=HySdkMemoryRecord,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def update_sdk_memory(
         memory_id: str,
@@ -3153,6 +3398,13 @@ def create_app(
         namespace = _namespace_kwargs(request.kind, request.variant)
         try:
             async with app.state.sdk_write_coordinator.use_client() as client:
+                # `update` takes no user: read the memory first and refuse
+                # another user's id exactly as a missing one.
+                current = await asyncio.to_thread(client.get, memory_id, **namespace)
+                if current is None or not owns(
+                    HySdkMemoryRecord.from_sdk(current).user_id, request.user_id
+                ):
+                    raise HTTPException(status_code=404, detail="memory_not_found")
                 update_kwargs: dict[str, object] = dict(namespace)
                 if request.task is not None:
                     update_kwargs["task"] = request.task
@@ -3190,16 +3442,27 @@ def create_app(
     @app.delete(
         "/api/v2/memories/{memory_id}",
         status_code=204,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def delete_sdk_memory(
         memory_id: str,
         kind: SdkMemoryKind | None = None,
         variant: SdkMemoryVariant | None = None,
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
     ) -> Response:
         _check_memory_namespace(kind, variant)
         try:
             async with app.state.sdk_write_coordinator.use_client() as client:
+                # `delete` takes no user. Another user's memory is answered
+                # exactly like an id that does not exist: 204, nothing
+                # touched, so the status cannot tell that it exists.
+                current = await asyncio.to_thread(
+                    client.get, memory_id, **_namespace_kwargs(kind, variant)
+                )
+                if current is not None and not owns(
+                    HySdkMemoryRecord.from_sdk(current).user_id, user_id
+                ):
+                    return Response(status_code=204)
                 result = await asyncio.to_thread(
                     client.delete, memory_id, **_namespace_kwargs(kind, variant)
                 )
@@ -3210,6 +3473,8 @@ def create_app(
                 raise HySdkError(str(error_code or "hy_memory_delete_incomplete"))
             if result.get("success") is not True and result.get("deleted_count") != 0:
                 raise HySdkError("hy_memory_delete_rejected")
+        except HTTPException:
+            raise
         except AmbiguousMemoryIdError:
             raise HTTPException(status_code=409, detail="memory_id_ambiguous") from None
         except Exception as error:
@@ -3219,18 +3484,19 @@ def create_app(
     @app.get(
         "/api/v2/raw",
         response_model=SdkRawPage,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def sdk_raw_library(
         source: str | None = Query(default=None, min_length=1),
         offset: int = Query(default=0, ge=0),
         limit: int = Query(default=100, ge=1, le=500),
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
     ) -> SdkRawPage:
         try:
             client = await _sdk_client_handle(app).client()
             value = await asyncio.to_thread(
                 client.list_memories,
-                user_id="local",
+                user_id=user_id,
                 layers=["l1_raw"],
                 include_sources=[source] if source else None,
                 offset=offset,
@@ -3244,7 +3510,7 @@ def create_app(
     @app.post(
         "/api/v2/memories/ask",
         response_model=SdkMemoryAnswer,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def ask_sdk_memories(
         request: AskMemoryRequest,
@@ -3261,7 +3527,9 @@ def create_app(
             "om_ask",
         )
         try:
-            answer = await _sdk_request_answerer(app).answer(request.question)
+            answer = await _sdk_request_answerer(app).answer(
+                request.question, user_id=request.user_id
+            )
         except MemoryAnswerError as error:
             raise HTTPException(status_code=503, detail=str(error)) from None
         if request.question != "__om_probe__" and access is not None:
@@ -3278,7 +3546,7 @@ def create_app(
 
     @app.post(
         "/api/v2/memories/ask/stream",
-        dependencies=protected,
+        dependencies=strict,
     )
     async def ask_sdk_memories_stream(
         request: AskMemoryRequest,
@@ -3300,7 +3568,9 @@ def create_app(
 
             final: dict[str, object] | None = None
             try:
-                async for event in answerer.answer_stream(request.question):
+                async for event in answerer.answer_stream(
+                    request.question, user_id=request.user_id
+                ):
                     if event.get("type") == "final":
                         final = event
                     yield _sse(_serialize_ask_event(event))
@@ -3331,7 +3601,7 @@ def create_app(
     @app.post(
         "/api/v1/recall",
         response_model=RecallReceipt,
-        dependencies=protected,
+        dependencies=strict,
     )
     async def recall_memories(
         request: RecallRequest,
@@ -3344,16 +3614,20 @@ def create_app(
         await _require_external_recall_scope(
             app, http_request, request.scope, request.workspace_path
         )
-        target = _recall_target(app)
+        target = (
+            _recall_target(app)
+            if request.user_id == DEFAULT_USER_ID
+            else _sdk_recall_target(app, request.user_id)
+        )
         access = external_access(http_request.headers)
         receipt = await recall_with_fallback(
             request,
             model_ready=lambda: _recall_model_ready(app),
             recall_memories=target.recall_async,
-            has_memories=lambda: _library_has_memories(app),
+            has_memories=lambda: _library_has_memories(app, request.user_id),
             raw_sources=lambda: _recall_raw_sources(app, request, access, started),
             search_raw=lambda query, limit, sources: _search_sdk_raw(
-                app, query, limit=limit, sources=sources
+                app, query, limit=limit, sources=sources, user_id=request.user_id
             ),
         )
         event_store = getattr(app.state, "call_event_store", None)
@@ -3730,21 +4004,60 @@ def _sync_refused(code: str) -> HTTPException:
     return HTTPException(status_code=409, detail=code)
 
 
+def _lane_sync_enabled(config_store: ConfigStore, lane: str) -> bool:
+    """Whether a lane may sync, before its grant is asked.
+
+    `local` keeps the App's per-source switch in the configuration document,
+    which stays global. Another user's Hermes grant syncs for as long as it
+    stands: connecting it is the consent, revoking it stops it, and it never
+    turns the App's switch on or off.
+    """
+    user_id, source_id = split_lane(lane)
+    if user_id == DEFAULT_USER_ID:
+        return config_store.sync_enabled(source_id)
+    return source_id in MULTI_USER_SOURCES
+
+
+def _other_user_lanes(
+    catalog: dict[str, SourceRegistration],
+    grant_store: SourceGrantStore,
+) -> tuple[str, ...]:
+    """Lanes of every unrevoked grant held by a user other than `local`."""
+    return tuple(
+        lane_key(receipt.user_id, receipt.source_id)
+        for receipt, _root in grant_store.grants()
+        if receipt.user_id != DEFAULT_USER_ID
+        and receipt.revoked_at is None
+        and receipt.source_id in MULTI_USER_SOURCES
+        and receipt.source_id in catalog
+    )
+
+
+def _known_user_ids(app: FastAPI) -> tuple[str, ...]:
+    """`local` and every user holding a grant: the users whose data Core keeps."""
+    users = [DEFAULT_USER_ID]
+    for receipt, _root in _grant_store(app).grants():
+        if receipt.user_id not in users:
+            users.append(receipt.user_id)
+    return tuple(users)
+
+
 def _granted_roots(
     source_ids: Iterable[str],
     grant_store: SourceGrantStore,
     config_store: ConfigStore,
 ) -> dict[str, Path]:
     roots: dict[str, Path] = {}
-    for source_id in source_ids:
-        if not config_store.sync_enabled(source_id):
+    for lane in source_ids:
+        if not _lane_sync_enabled(config_store, lane):
             continue
+        user_id, source_id = split_lane(lane)
         try:
-            grant = grant_store.authorized_root(user_id="local", source_id=source_id)
+            grant = grant_store.authorized_root(user_id=user_id, source_id=source_id)
         except SourceAccessError:
             continue
         if grant is not None:
-            roots[source_id] = grant.root
+            roots[lane] = grant.root
     return roots
 
 
@@ -3753,6 +4066,14 @@ def _active_adapters(
     grant_store: SourceGrantStore,
 ) -> dict[str, SourceAdapter]:
     adapters: dict[str, SourceAdapter] = {}
+    for lane in _other_user_lanes(catalog, grant_store):
+        user_id, source_id = split_lane(lane)
+        try:
+            grant = grant_store.authorized_root(user_id=user_id, source_id=source_id)
+            if grant is not None:
+                adapters[lane] = catalog[source_id].create_adapter(grant)
+        except SourceAccessError:
+            continue
     for source_id, registration in catalog.items():
         try:
             grant = grant_store.authorized_root(
@@ -3915,10 +4236,14 @@ def _config_store(app: FastAPI) -> ConfigStore:
 
 
 def _source_processing_enabled(app: FastAPI, source_id: str) -> bool:
-    if not _config_store(app).sync_enabled(source_id):
+    """`source_id` is a lane key: another user's lane asks that user's grant."""
+    if not _lane_sync_enabled(_config_store(app), source_id):
         return False
+    user_id, bare_source_id = split_lane(source_id)
     try:
-        return _grant_store(app).authorized_root(user_id="local", source_id=source_id) is not None
+        return (
+            _grant_store(app).authorized_root(user_id=user_id, source_id=bare_source_id) is not None
+        )
     except SourceAccessError:
         return False
 
@@ -4437,13 +4762,32 @@ async def _recall_model_ready(app: FastAPI) -> bool:
     return profile is not None and bool(profile.capabilities.model_available)
 
 
-async def _library_has_memories(app: FastAPI) -> bool:
+def _sdk_recall_target(app: FastAPI, user_id: str) -> SdkRecallTarget:
+    """Recall over one user's SDK memories; `get` results are kept only if theirs."""
+    handle = _sdk_client_handle(app)
+
+    async def owned(memory_id: str) -> HySdkMemoryRecord | None:
+        memory = await _get_sdk_memory(handle, memory_id)
+        return memory if memory is not None and owns(memory.user_id, user_id) else None
+
+    return SdkRecallTarget(
+        semantic_search=lambda query: _search_sdk_memories(
+            handle,
+            query,
+            user_id=user_id,
+            agent_id="*",
+        ),
+        get_memory=owned,
+    )
+
+
+async def _library_has_memories(app: FastAPI, user_id: str = DEFAULT_USER_ID) -> bool:
     """Any active memory at all. Asked only after an empty memory search."""
     try:
         client = await _sdk_client_handle(app).client()
         raw = await asyncio.to_thread(
             client.list_memories,
-            user_id="local",
+            user_id=user_id,
             statuses=["active"],
             limit=1,
             offset=0,
@@ -4501,13 +4845,14 @@ async def _search_sdk_raw(
     *,
     limit: int,
     sources: frozenset[str] | None,
+    user_id: str = DEFAULT_USER_ID,
 ) -> tuple[HySdkRawHit, ...]:
     """The imported conversations nearest the query: the SDK raw layer."""
     client = await _sdk_client_handle(app).client()
     raw = await asyncio.to_thread(
         client.search,
         query,
-        user_ids=["local"],
+        user_ids=[user_id],
         agent_ids=[],
         include_sources=sorted(sources) if sources else None,
         limit=limit,
@@ -4610,6 +4955,7 @@ async def _library_search_page(
     limit: int,
     window_id: str | None,
     scene: str = "all",
+    user_id: str = DEFAULT_USER_ID,
 ) -> HySdkMemoryPage:
     """One page of a keyword search, cut from a cached result window.
 
@@ -4621,7 +4967,9 @@ async def _library_search_page(
     SDK so edits and deletes show. The whole library is never listed in Core.
     """
     windows = _library_search_windows(app)
-    key = window_key(query, source, statuses, scene)
+    # The user is part of the window's identity: one user's `window_id` must
+    # never cut a page from another user's matches.
+    key = window_key(query, source, statuses, scene, user_id)
     window = windows.get(window_id, key) if window_id else None
     window_reset = window_id is not None and window is None
     if window is None:
@@ -4631,7 +4979,7 @@ async def _library_search_page(
             query,
             scene="all",
             rerank=False,
-            user_ids=["local"],
+            user_ids=[user_id],
             agent_ids=[],
             include_sources=[source] if source else None,
             limit=fetch,
@@ -4667,7 +5015,7 @@ async def _library_search_page(
         except HySdkReadError:
             skipped += 1
             continue
-        if record.status == "active":
+        if record.status == "active" and owns(record.user_id, user_id):
             memories.append(record)
     _log_skipped_sdk_memories(skipped)
     return HySdkMemoryPage(
@@ -4724,8 +5072,10 @@ def _source_writes_on(app: FastAPI, source_id: str) -> bool:
     forever with attempts=0 while later jobs ran. Intake, claim and execute
     all ask this one question.
     """
-    if source_id == "one-memory-manual":
-        receipt = _grant_store(app).receipt(user_id="local", source_id=source_id)
+    if split_lane(source_id)[1] == "one-memory-manual":
+        # One switch for hand-added memories, whoever adds them: it is a
+        # management setting, and management stays global.
+        receipt = _grant_store(app).receipt(user_id="local", source_id="one-memory-manual")
         return receipt is None or receipt.revoked_at is None
     return _source_processing_enabled(app, source_id)
 
@@ -4759,6 +5109,7 @@ def _history_import_gate_released(app: FastAPI) -> bool:
         job
         for job in jobs.list_jobs()
         if job.operation == "history"
+        and job.user_id == DEFAULT_USER_ID
         and job.source_id in selected
         and (state.started_at is None or job.created_at >= state.started_at)
     ]

memhub/apps/core/src/one_memory/context_models.py

@@ -5,6 +5,8 @@ from typing import Annotated, Literal
 
 from pydantic import BaseModel, ConfigDict, Field, model_validator
 
+from one_memory.user_scope import DEFAULT_USER_ID, UserId
+
 ContextKind = Literal["raw", "memory"]
 ContextSearchChannel = Literal["lexical", "semantic"]
 ContextSemanticStatus = Literal["not_searched", "completed", "failed", "timed_out"]
@@ -58,6 +60,8 @@ class SdkContextSearchRequest(ContextSearchRequest):
     limit: int = Field(default=10, ge=1, le=20, strict=True)
     schema_limit: int = Field(default=3, ge=0, le=20, strict=True)
     intention_limit: int = Field(default=1, ge=0, le=20, strict=True)
+    #: Whose memories to search. The App never sends it and reads `local`.
+    user_id: UserId = DEFAULT_USER_ID
 
 
 class ContextMessage(BaseModel):

memhub/apps/core/src/one_memory/context_routes.py

@@ -28,6 +28,7 @@ from one_memory.hy_sdk_types import (
     parse_sdk_search,
 )
 from one_memory.imports.storage import EvidenceStore
+from one_memory.user_scope import owns
 
 CALL_EVENT_RESPONSE_BUDGET_MS = 250
 
@@ -73,7 +74,7 @@ def sdk_context_router(dependencies: Sequence[Depends]) -> APIRouter:
             raw = await asyncio.to_thread(
                 client.search,
                 value.query,
-                user_ids=["local"],
+                user_ids=[value.user_id],
                 agent_ids=[],
                 include_sources=list(authority.raw_sources or ()),
                 limit=value.limit,
@@ -108,7 +109,7 @@ def sdk_context_router(dependencies: Sequence[Depends]) -> APIRouter:
             raw = await asyncio.to_thread(
                 client.search,
                 value.query,
-                user_ids=["local"],
+                user_ids=[value.user_id],
                 agent_ids=[],
                 limit=value.limit,
                 min_score=sdk_relevance_floor(),
@@ -123,7 +124,12 @@ def sdk_context_router(dependencies: Sequence[Depends]) -> APIRouter:
             for hit in hits:
                 record = await asyncio.to_thread(client.get, hit.memory_id)
                 memory = None if record is None else HySdkMemoryRecord.from_sdk(record)
-                if memory is None or memory.status != "active":
+                # `get` takes no user: a hit is kept only if it is the caller's.
+                if (
+                    memory is None
+                    or memory.status != "active"
+                    or not owns(memory.user_id, value.user_id)
+                ):
                     continue
                 items.append(
                     SdkMemoryContextItem(

memhub/apps/core/src/one_memory/hy_runtime_config.py

@@ -14,7 +14,7 @@ from pathlib import Path
 from typing import Literal, cast
 
 HY_MEMORY_DISTRIBUTION = "hy-memory"
-HY_MEMORY_VERSION = "2.0.8"
+HY_MEMORY_VERSION = "2.0.9"
 HY_MEMORY_NAMESPACE = "governed-v2"
 DiagnosticLogLevel = Literal["error", "warning", "info", "debug"]
 

memhub/apps/core/src/one_memory/hy_sdk_migration.py

@@ -497,7 +497,7 @@ def _ensure_coding_schema(path: Path) -> None:
         CodingMemoryStore,
     )
 
-    # Relies on the pinned SDK's internals (hy-memory==2.0.8 is pinned exactly).
+    # Relies on the pinned SDK's internals (hy-memory==2.0.9 is pinned exactly).
     store = CodingMemoryStore.__new__(CodingMemoryStore)
     store._db_path = str(path)
     store._conn = None

memhub/apps/core/src/one_memory/imports/adapters.py

@@ -31,6 +31,16 @@ class ScanIssueCode(StrEnum):
     #: import it. Named for the reason, so a reader can tell a setting from
     #: a record the parser could not use.
     SUBAGENT_TRACE_DISABLED = "subagent_trace_disabled"
+    #: A host's own notice wearing the user's role — Hermes posts a finished
+    #: background process, and its gateway marks the post as not written by a
+    #: human. Nothing in it is the person's, so it never reaches extraction.
+    INJECTED_MACHINE_NOTICE = "injected_machine_notice"
+    #: The host could not transcribe a voice message and left a placeholder
+    #: naming the audio file. The person did speak, but the placeholder holds
+    #: only a path, which an extractor would otherwise read as a fact.
+    VOICE_PLACEHOLDER = "voice_placeholder"
+    #: A health check such as "Reply with exactly: PONG": no one's words.
+    PROBE = "probe"
     SOURCE_BUSY = "source_busy"
     UNSUPPORTED_SCHEMA = "unsupported_schema"
     BUDGET_EXCEEDED = "budget_exceeded"

memhub/apps/core/src/one_memory/imports/discovery.py

@@ -27,6 +27,7 @@ def discovered_off_loop(
     *,
     suffixes: frozenset[str],
     max_targets: int,
+    excluded_top_directories: frozenset[str] = frozenset(),
 ) -> AsyncIterator[SourceFileDiscoveryEvent]:
     """The same walk, for callers that have an event loop to keep free.
 
@@ -41,6 +42,7 @@ def discovered_off_loop(
             grant,
             suffixes=suffixes,
             max_targets=max_targets,
+            excluded_top_directories=excluded_top_directories,
         ),
         scope_key=grant.source_id,
     )
@@ -51,12 +53,15 @@ def discover_source_files(
     *,
     suffixes: frozenset[str],
     max_targets: int,
+    excluded_top_directories: frozenset[str] = frozenset(),
 ) -> Iterator[SourceFileDiscoveryEvent]:
     """Metadata only: one lstat per candidate, no open and no per-file grant lookup.
 
     The grant and root identity are checked once before the walk. Each listed
     file keeps the grant's check, which runs again when that file is opened;
     the open itself refuses symlinks and any inode other than the one listed.
+    `excluded_top_directories` are directories directly under the root that
+    are never entered, such as another user's Hermes profiles.
     """
     grant.check_root()
     inspected = 0
@@ -64,8 +69,12 @@ def discover_source_files(
         grant.root,
         followlinks=False,
     ):
+        at_root = Path(directory) == grant.root
         child_directories[:] = sorted(
-            child for child in child_directories if not (Path(directory) / child).is_symlink()
+            child
+            for child in child_directories
+            if not (Path(directory) / child).is_symlink()
+            and not (at_root and child in excluded_top_directories)
         )
         for filename in sorted(filenames):
             if Path(filename).suffix.lower() not in suffixes:

memhub/apps/core/src/one_memory/imports/grants.py

@@ -252,6 +252,18 @@ class SourceGrantStore:
             ).fetchone()
         return self._to_receipt(row) if row else None
 
+    def grants(self) -> tuple[tuple[SourceGrantReceipt, str], ...]:
+        """Every grant of every user, with its granted root path.
+
+        One source may hold several grants, one per `(user_id, source_id)`:
+        a Hermes root for `local` and a Hermes profile for another user.
+        """
+        with self._connect() as connection:
+            rows = connection.execute(
+                "SELECT * FROM source_grants ORDER BY source_id, user_id"
+            ).fetchall()
+        return tuple((self._to_receipt(row), str(row["requested_path"])) for row in rows)
+
     def require_receipt(self, *, user_id: str, source_id: str) -> SourceGrantReceipt:
         receipt = self.receipt(user_id=user_id, source_id=source_id)
         if receipt is None:

memhub/apps/core/src/one_memory/imports/live_sync.py

@@ -28,6 +28,7 @@ from one_memory.imports.adapters import (
 )
 from one_memory.imports.jobs import ImportJob, ImportJobStore, ImportStatus
 from one_memory.imports.supervisor import ImportSupervisor
+from one_memory.user_scope import DEFAULT_USER_ID, lane_key, split_lane
 
 #: How long an automatic retry will wait for a wedged worker to notice cancel.
 ABANDON_TIMEOUT_SECONDS = 5.0
@@ -64,7 +65,12 @@ class LiveSyncHealth(BaseModel):
 
 
 class PollingLiveSyncSupervisor:
-    """Near-live sync that reuses each adapter's persisted incremental cursor."""
+    """Near-live sync that reuses each adapter's persisted incremental cursor.
+
+    A "source" here is a lane key (`user_scope.lane_key`): the bare source id
+    for `local`, `source@user` for another user's grant. Cursors, jobs and
+    locks are therefore per `(user_id, source_id)`.
+    """
 
     def __init__(
         self,
@@ -92,8 +98,22 @@ class PollingLiveSyncSupervisor:
         }
         self._locks = {source_id: asyncio.Lock() for source_id in self.source_ids}
         self._task: asyncio.Task[None] | None = None
+        self._group: asyncio.TaskGroup | None = None
         self._observers: set[asyncio.Task[None]] = set()
 
+    def ensure_source(self, source_id: str) -> None:
+        """Follow a lane granted after start, such as a new Hermes profile."""
+        if source_id in self._locks:
+            return
+        self.source_ids = (*self.source_ids, source_id)
+        self._health.setdefault(
+            source_id,
+            LiveSyncHealth(source_id=source_id, mode=self.mode, state=LiveSyncState.IDLE),
+        )
+        self._locks[source_id] = asyncio.Lock()
+        if self._group is not None:
+            self._group.create_task(self._poll_source(source_id))
+
     def _set_health(self, source_id: str, value: LiveSyncHealth) -> None:
         previous = self._health.get(source_id)
         self._health[source_id] = value
@@ -152,7 +172,12 @@ class PollingLiveSyncSupervisor:
         self._watched_sources = frozenset(source_ids)
 
     def all_health(self) -> tuple[LiveSyncHealth, ...]:
-        return tuple(self.health(source_id) for source_id in self.source_ids)
+        """`local`'s lanes only: what the App has always been shown."""
+        return tuple(
+            self.health(source_id)
+            for source_id in self.source_ids
+            if split_lane(source_id)[0] == DEFAULT_USER_ID
+        )
 
     async def initialize_source(
         self, source_id: str, *, baseline_missing_files: bool = True
@@ -164,11 +189,12 @@ class PollingLiveSyncSupervisor:
         baselines = (
             await adapter.live_baseline() if isinstance(adapter, _LiveBaselineAdapter) else {}
         )
+        user_id, bare_source_id = split_lane(source_id)
         await in_the_background(
             partial(
                 self.job_store.ensure_live_cursors,
-                user_id="local",
-                source_id=source_id,
+                user_id=user_id,
+                source_id=bare_source_id,
                 baselines=baselines,
                 baseline_missing_files=baseline_missing_files,
             )
@@ -228,7 +254,7 @@ class PollingLiveSyncSupervisor:
                 (
                     job
                     for job in reversed(source_jobs)
-                    if job.source_id == source_id
+                    if lane_key(job.user_id, job.source_id) == source_id
                     and job.operation == "sync"
                     and job.status
                     in {
@@ -277,13 +303,13 @@ class PollingLiveSyncSupervisor:
             competing = [
                 job
                 for job in source_jobs
-                if job.source_id == source_id
+                if lane_key(job.user_id, job.source_id) == source_id
                 and job.operation != "sync"
                 and job.status
                 in {ImportStatus.PENDING, ImportStatus.RUNNING, ImportStatus.INTERRUPTED}
             ]
             bounded_history_incomplete = any(
-                job.source_id == source_id
+                lane_key(job.user_id, job.source_id) == source_id
                 and job.operation == "history"
                 and job.history_after is not None
                 and job.status is not ImportStatus.COMPLETED
@@ -379,11 +405,12 @@ class PollingLiveSyncSupervisor:
                 await close()
 
     async def _live_cursors(self, source_id: str, adapter: object) -> dict[str, SourceCursor]:
+        user_id, bare_source_id = split_lane(source_id)
         cursors = await in_the_background(
             partial(
                 self.job_store.get_source_cursors,
-                user_id="local",
-                source_id=source_id,
+                user_id=user_id,
+                source_id=bare_source_id,
                 operation="sync",
             )
         )
@@ -395,8 +422,8 @@ class PollingLiveSyncSupervisor:
         return await in_the_background(
             partial(
                 self.job_store.ensure_live_cursors,
-                user_id="local",
-                source_id=source_id,
+                user_id=user_id,
+                source_id=bare_source_id,
                 baselines=baselines,
             )
         )
@@ -520,9 +547,13 @@ class PollingLiveSyncSupervisor:
     async def _run(self) -> None:
         # A source owns its cadence; one large catch-up cannot delay every
         # other source. The group keeps shutdown responsible for every poller.
-        async with asyncio.TaskGroup() as group:
-            for source_id in self.source_ids:
-                group.create_task(self._poll_source(source_id))
+        try:
+            async with asyncio.TaskGroup() as group:
+                self._group = group
+                for source_id in self.source_ids:
+                    group.create_task(self._poll_source(source_id))
+        finally:
+            self._group = None
 
     async def _poll_source(self, source_id: str) -> None:
         while True:

memhub/apps/core/src/one_memory/imports/runner.py

@@ -3,7 +3,7 @@ import logging
 import math
 import sqlite3
 from collections import Counter
-from collections.abc import Callable
+from collections.abc import AsyncIterator, Callable
 from contextlib import suppress
 from contextvars import ContextVar
 from dataclasses import dataclass, field
@@ -227,14 +227,17 @@ class ImportRunner:
         heartbeat = asyncio.create_task(self._keep_alive(job_id=job_id, runner_token=runner_token))
         pending = _PendingSkips()
         writes = _PendingWrites()
-        stream = adapter.scan(
-            ScanOptions(
-                cursors=cursors,
-                reuse_listing=(
-                    job.operation == "history"
-                    or (job.operation == "sync" and job.source_id == "workbuddy")
-                ),
-            )
+        stream = _owned_by(
+            adapter.scan(
+                ScanOptions(
+                    cursors=cursors,
+                    reuse_listing=(
+                        job.operation == "history"
+                        or (job.operation == "sync" and job.source_id == "workbuddy")
+                    ),
+                )
+            ),
+            job.user_id,
         )
         slice_cursors = dict(cursors)
         next_event: asyncio.Future[ScanEvent] | None = None
@@ -924,3 +927,30 @@ async def _settled_work[T](
         with suppress(asyncio.CancelledError, Exception):
             task.result()
         raise
+
+
+def _owned_by(stream: AsyncIterator[ScanEvent], user_id: str) -> AsyncIterator[ScanEvent]:
+    """Stamp every scanned record with the user whose root is being read.
+
+    Adapters know nothing of users and emit `local`. Which user a record
+    belongs to is decided by the grant the job reads (a Hermes profile's
+    `state.db` is that profile's user), so the job's user is the record's.
+    `local` passes the stream through untouched.
+    """
+    if user_id == "local":
+        return stream
+    return _restamped(stream, user_id)
+
+
+async def _restamped(stream: AsyncIterator[ScanEvent], user_id: str) -> AsyncIterator[ScanEvent]:
+    try:
+        async for event in stream:
+            if isinstance(event, ScannedEvidence) and event.evidence.user_id != user_id:
+                event = event.model_copy(
+                    update={"evidence": event.evidence.model_copy(update={"user_id": user_id})}
+                )
+            yield event
+    finally:
+        close = getattr(stream, "aclose", None)
+        if close is not None:
+            await close()

memhub/apps/core/src/one_memory/imports/sources/hermes.py

@@ -1,4 +1,5 @@
 import json
+import re
 import sqlite3
 from collections.abc import AsyncIterator, Iterator
 from functools import partial
@@ -87,6 +88,7 @@ class HermesSourceAdapter:
             self.grant,
             suffixes=frozenset({".jsonl", ".db"}),
             max_targets=10_000,
+            excluded_top_directories=frozenset({HERMES_PROFILES_DIRECTORY}),
         ):
             if isinstance(item, AuthorizedSourceFile) and _is_hermes_target(
                 item.relative_path.as_posix()
@@ -100,6 +102,7 @@ class HermesSourceAdapter:
             self.grant,
             suffixes=frozenset({".jsonl", ".db"}),
             max_targets=options.max_scan_targets or 100_000,
+            excluded_top_directories=frozenset({HERMES_PROFILES_DIRECTORY}),
         ):
             yield ScanProgressEvent(
                 phase=ScanPhase.DISCOVER,
@@ -214,7 +217,11 @@ def parse_hermes_rollout_record(
         message.get("content"),
         allowed_block_types=frozenset({"text", "input_text", "output_text"}),
     )
-    text = strip_injected_memory(text)
+    text = strip_injected_memory(text or "")
+    if role == "user":
+        text, injected = gateway_user_text(text)
+        if injected is not None:
+            return ParsedSkip(reason_code=injected, conversation_id=conversation_id)
     if not text:
         return ParsedSkip(
             reason_code=ScanIssueCode.NO_TEXT,
@@ -242,8 +249,17 @@ def parse_hermes_rollout_record(
     )
 
 
+#: Hermes keeps each extra profile under `<root>/profiles/<name>/`, with its
+#: own `state.db` and sessions. A profile is another person's history: it is
+#: read only through its own grant (its own MemHub user), never as part of the
+#: root above it. Hermes' own export skips this directory for the same reason.
+HERMES_PROFILES_DIRECTORY = "profiles"
+
+
 def _is_hermes_target(relative_path: str) -> bool:
     path = PurePosixPath(relative_path)
+    if path.parts and path.parts[0] == HERMES_PROFILES_DIRECTORY:
+        return False
     return path.name == "state.db" or (path.suffix == ".jsonl" and "sessions" in path.parts)
 
 
@@ -389,6 +405,15 @@ def _read_state_db(
             # `tool` rows after it hold only what came back.
             calls = _tool_calls_text(row["tool_calls"]) if role == "assistant" else None
             content = strip_injected_memory(str(row["content"] or ""))
+            if role == "user":
+                content, injected = gateway_user_text(content)
+                if injected is not None:
+                    yield ScanSkip(
+                        cursor=cursor,
+                        conversation_id=str(row["session_id"]),
+                        reason_code=injected,
+                    )
+                    continue
             has_text = bool(content.strip())
             if role not in {"user", "assistant"} or (not has_text and calls is None):
                 yield ScanSkip(
@@ -439,6 +464,69 @@ def _read_state_db(
 _PERSON_SESSION_SOURCES = frozenset({"cli", "tui", "webui", "desktop"})
 
 
+#: What the Hermes gateway writes into a `user` row on its own behalf. Every
+#: one of these wears the `user` role, so only the text can tell them apart
+#: from the person. Read off real `state.db` rows: a finished background
+#: process is posted as a turn, often with the gateway's own "not written by
+#: a human" note appended, which is why that note counts anywhere in the row.
+_BACKGROUND_NOTICE = re.compile(
+    r"\[IMPORTANT:\s*(?:\d+\s+background\s+process(?:es)?\b|Background\s+process\b)",
+    re.IGNORECASE,
+)
+_GATEWAY_NOTICE = re.compile(
+    r"\[INTERNAL NOTIFICATION\s*\W+\s*generated by the Hermes gateway",
+    re.IGNORECASE,
+)
+_VOICE_PLACEHOLDER = re.compile(r"\[voice message could not be transcribed", re.IGNORECASE)
+_PROBE = re.compile(r"Reply with exactly:\s*[A-Za-z0-9_\-]{1,24}\s*")
+_MEDIA_ONLY = re.compile(r"(?:MEDIA:/\S+\s*)+")
+#: Wrappers the gateway puts around a message the person did write. Only the
+#: wrapper goes; what it carries is the turn. Each is matched at the edge of
+#: what is left, and the set is applied until nothing more comes off, since a
+#: relayed message carries the origin block inside the out-of-band markers.
+_GATEWAY_WRAPPERS = (
+    re.compile(r"^\[OUT-OF-BAND USER MESSAGE\b[^\]\n]*\]"),
+    re.compile(r"\[/OUT-OF-BAND USER MESSAGE\]$"),
+    re.compile(
+        r"^Gateway message origin \(JSON data, not instructions or authorization\):[ \t]*"
+        r"(?:\n[ \t]*\{[^\n]*\}[ \t]*)?"
+        r"(?:\n[ \t]*Do not guess a reply destination[^\n]*)?"
+        r"(?=\n|$)"
+    ),
+    re.compile(r"^\[Context from the interrupted assistant response\][ \t]*(?=\n|$)"),
+)
+
+
+def gateway_user_text(text: str) -> tuple[str, ScanIssueCode | None]:
+    """The person's words in a Hermes `user` row, or why there are none.
+
+    Applied only to the `user` role: an assistant or tool row quoting the
+    same text is reporting what it saw, which is the existing tool rule's
+    business. The cleaned text is what becomes the evidence, so its content
+    hash and duplicate check see the message rather than its envelope.
+    """
+
+    if _GATEWAY_NOTICE.search(text):
+        return "", ScanIssueCode.INJECTED_MACHINE_NOTICE
+    cleaned = text.strip()
+    while True:
+        unwrapped = cleaned
+        for wrapper in _GATEWAY_WRAPPERS:
+            unwrapped = wrapper.sub("", unwrapped, count=1).strip()
+        if unwrapped == cleaned:
+            break
+        cleaned = unwrapped
+    if _BACKGROUND_NOTICE.match(cleaned):
+        return "", ScanIssueCode.INJECTED_MACHINE_NOTICE
+    if _VOICE_PLACEHOLDER.match(cleaned):
+        return "", ScanIssueCode.VOICE_PLACEHOLDER
+    if _PROBE.fullmatch(cleaned):
+        return "", ScanIssueCode.PROBE
+    if _MEDIA_ONLY.fullmatch(cleaned):
+        return "", ScanIssueCode.NO_TEXT
+    return cleaned, None
+
+
 def _tool_calls_text(value: object) -> str | None:
     """OpenAI-shaped `tool_calls`: `[{"function": {"name", "arguments"}}]`, stored as JSON."""
     if isinstance(value, str):

memhub/apps/core/src/one_memory/imports/supervisor.py

@@ -24,6 +24,7 @@ from one_memory.service_levels import (
     ServiceLevelScheduler,
     ServiceResource,
 )
+from one_memory.user_scope import lane_key, split_lane
 
 
 class UnknownSourceError(KeyError):
@@ -82,36 +83,45 @@ class ImportSupervisor:
         self._tasks: dict[str, asyncio.Task[None]] = {}
         self._drains: dict[str, asyncio.Task[None]] = {}
 
-    def register_adapter(self, adapter: SourceAdapter) -> None:
-        self.adapters[adapter.descriptor.source_id] = adapter
+    def register_adapter(self, adapter: SourceAdapter, *, user_id: str = "local") -> None:
+        """Adapters are keyed by lane: the source id for `local`, `source@user`
+        for another user's grant of the same source."""
+        self.adapters[lane_key(user_id, adapter.descriptor.source_id)] = adapter
 
     def unregister_adapter(self, source_id: str) -> None:
         self.adapters.pop(source_id, None)
 
+    @staticmethod
+    def lane_of(job: ImportJob) -> str:
+        return lane_key(job.user_id, job.source_id)
+
     def active_job_ids(self) -> frozenset[str]:
         return frozenset(job_id for job_id, task in self._tasks.items() if not task.done())
 
     def create_job(
         self, source_id: str, *, operation: Literal["history", "sync"] = "history"
     ) -> ImportJob:
+        """`source_id` is a lane key: another user's lane creates that user's job."""
         if self.source_enabled is not None and not self.source_enabled(source_id):
             raise SourceAccessError(SourceAccessErrorCode.GRANT_REQUIRED)
-        return self.job_store.create(user_id="local", source_id=source_id, operation=operation)
+        user_id, bare_source_id = split_lane(source_id)
+        return self.job_store.create(user_id=user_id, source_id=bare_source_id, operation=operation)
 
     def enqueue(self, job_id: str) -> ImportJob:
         if job_id in self._drains:
             raise InvalidJobTransitionError("import_cleanup_in_progress")
         job = self.job_store.require(job_id)
-        if job.source_id not in self.adapters:
-            raise UnknownSourceError(job.source_id)
-        if self.source_enabled is not None and not self.source_enabled(job.source_id):
+        lane = self.lane_of(job)
+        if lane not in self.adapters:
+            raise UnknownSourceError(lane)
+        if self.source_enabled is not None and not self.source_enabled(lane):
             return job
         existing = self._tasks.get(job_id)
         if existing is None or existing.done():
             self._tasks[job_id] = asyncio.create_task(
                 self._execute(
                     job_id,
-                    source_id=job.source_id,
+                    source_id=lane,
                     operation=job.operation,
                 )
             )
@@ -130,8 +140,10 @@ class ImportSupervisor:
         # pending history. Each resumed runner will yield after one bounded slice.
         for status in (ImportStatus.INTERRUPTED, ImportStatus.PENDING):
             for job in self.job_store.list_jobs(status=status):
-                if job.source_id in self.adapters:
-                    if self.source_enabled is not None and not self.source_enabled(job.source_id):
+                if self.lane_of(job) in self.adapters:
+                    if self.source_enabled is not None and not self.source_enabled(
+                        self.lane_of(job)
+                    ):
                         continue
                     if job.status is not ImportStatus.PENDING:
                         self.job_store.queue_for_resume(job.job_id)
@@ -227,7 +239,7 @@ class ImportSupervisor:
                     }:
                         return
                     if self.source_enabled is not None and not await in_the_background(
-                        self.source_enabled, job.source_id
+                        self.source_enabled, self.lane_of(job)
                     ):
                         return
                     runner = ImportRunner(
@@ -238,7 +250,7 @@ class ImportSupervisor:
                         sdk_queue=self.sdk_queue,
                         sdk_authority=self.sdk_authority,
                     )
-                    adapter = self.adapters[job.source_id]
+                    adapter = self.adapters[self.lane_of(job)]
                     if self.prepare_adapter is not None:
                         adapter = await self.prepare_adapter(job, adapter)
 
@@ -273,7 +285,7 @@ class ImportSupervisor:
                 and retry_attempt < self.max_history_retries
                 and (
                     self.source_enabled is None
-                    or await in_the_background(self.source_enabled, current.source_id)
+                    or await in_the_background(self.source_enabled, self.lane_of(current))
                 )
             ):
                 delay = self.retry_base_seconds * (2**retry_attempt)

memhub/apps/core/src/one_memory/library_search_window.py

@@ -7,9 +7,9 @@ between pages. The matches are kept here under an opaque ``window_id`` that
 the client passes back with the next offset.
 
 A window is bound to the filters it was searched with (query, source,
-statuses, scene). It expires after ``ttl_seconds`` without being read, and the
-store keeps at most ``max_windows`` windows, dropping the least recently read
-first.
+statuses, scene) and to the user whose library it searched. It expires after
+``ttl_seconds`` without being read, and the store keeps at most
+``max_windows`` windows, dropping the least recently read first.
 Only memory identities are kept, never content: each page is read fresh from
 the SDK, so an edit or a delete shows on the next page read.
 """
@@ -28,13 +28,17 @@ from one_memory.hy_sdk_types import LibraryHit
 DEFAULT_WINDOW_TTL_SECONDS = 300.0
 DEFAULT_MAX_WINDOWS = 32
 
-WindowKey = tuple[str, str | None, tuple[str, ...], str]
+WindowKey = tuple[str, str | None, tuple[str, ...], str, str]
 
 
 def window_key(
-    query: str, source: str | None, statuses: list[str], scene: str = "all"
+    query: str,
+    source: str | None,
+    statuses: list[str],
+    scene: str = "all",
+    user_id: str = "local",
 ) -> WindowKey:
-    return (query, source, tuple(sorted(set(statuses))), scene)
+    return (query, source, tuple(sorted(set(statuses))), scene, user_id)
 
 
 @dataclass(frozen=True)

memhub/apps/core/src/one_memory/memory_additions.py

@@ -11,10 +11,11 @@ import asyncio
 import sqlite3
 from collections.abc import Sequence
 from datetime import UTC, datetime
+from functools import partial
 from typing import Any, Literal, cast
 from uuid import UUID
 
-from fastapi import APIRouter, BackgroundTasks, HTTPException, Request
+from fastapi import APIRouter, BackgroundTasks, HTTPException, Query, Request
 from fastapi.params import Depends
 from pydantic import BaseModel, ConfigDict, Field, field_validator
 
@@ -34,6 +35,7 @@ from one_memory.memory_server.durable_queue import DurableWriteQueue
 from one_memory.memory_server.unified_entry import UnifiedWriteIntake
 from one_memory.memory_tasks import MemoryTaskStatus, MemoryTaskStore
 from one_memory.secrets import SecretStoreUnavailableError
+from one_memory.user_scope import DEFAULT_USER_ID, USER_ID_PATTERN, UserId
 
 SOURCE_ID = "one-memory-manual"
 MemoryAddState = Literal[
@@ -53,6 +55,8 @@ class MemoryAddInput(BaseModel):
     request_id: UUID
     text: str = Field(min_length=1, max_length=12_000)
     scope: Literal["all-apps"]
+    #: Whose memory this is. The App never sends it and writes `local`.
+    user_id: UserId = DEFAULT_USER_ID
 
     @field_validator("text")
     @classmethod
@@ -108,7 +112,11 @@ class MemoryAdditionStore:
 
     def save(self, addition: MemoryAddInput) -> EvidenceRecord:
         now = datetime.now(UTC)
+        if addition.user_id != DEFAULT_USER_ID and not self.sdk_authority:
+            # Legacy authority keeps one person's library only.
+            raise ValueError("memory_add_user_requires_sdk_authority")
         value = EvidenceInput(
+            user_id=addition.user_id,
             source_id=SOURCE_ID,
             conversation_id=f"note:{addition.request_id}",
             message_id=str(addition.request_id),
@@ -200,7 +208,7 @@ class MemoryAdditionStore:
                 )
         return self.require(value.evidence_id)
 
-    def require(self, evidence_id: str) -> EvidenceRecord:
+    def require(self, evidence_id: str, *, user_id: str = DEFAULT_USER_ID) -> EvidenceRecord:
         if self.sdk_authority:
             if self.write_queue is None:
                 raise KeyError(evidence_id)
@@ -219,11 +227,11 @@ class MemoryAdditionStore:
                 (item for item in payload.input_records if item.evidence_id == evidence_id),
                 None,
             )
-            if record is None:
+            if record is None or record.user_id != user_id:
                 raise KeyError(evidence_id)
             return record
         record = self.evidence.get(evidence_id)
-        if record is None or record.user_id != "local" or record.source_id != SOURCE_ID:
+        if record is None or record.user_id != user_id or record.source_id != SOURCE_ID:
             raise KeyError(evidence_id)
         metadata = record.origin_meta.get(EvidenceOrigin.MANUAL_IMPORT)
         if metadata is None or metadata.raw_meta.get("input_mode") != "direct_memory_note":
@@ -237,8 +245,8 @@ class MemoryAdditionStore:
             raise ValueError("memory_add_source_revoked")
         return record
 
-    def receipt(self, evidence_id: str) -> MemoryAddReceipt:
-        record = self.require(evidence_id)
+    def receipt(self, evidence_id: str, *, user_id: str = DEFAULT_USER_ID) -> MemoryAddReceipt:
+        record = self.require(evidence_id, user_id=user_id)
         if self.sdk_authority:
             assert self.write_queue is not None
             with main_database(self.write_queue.database_path).read() as database:
@@ -300,10 +308,15 @@ class MemoryAdditionStore:
             error=error,
         )
 
-    def receipt_for_request(self, request_id: UUID) -> MemoryAddReceipt:
+    def receipt_for_request(
+        self, request_id: UUID, *, user_id: str = DEFAULT_USER_ID
+    ) -> MemoryAddReceipt:
         identity = str(request_id)
         if self.sdk_authority:
+            # The Evidence id carries the user, so another user's request id
+            # names a different note and reads as not found.
             probe = EvidenceInput(
+                user_id=user_id,
                 source_id=SOURCE_ID,
                 conversation_id=f"note:{identity}",
                 message_id=identity,
@@ -316,16 +329,16 @@ class MemoryAdditionStore:
                     )
                 },
             )
-            return self.receipt(probe.evidence_id)
+            return self.receipt(probe.evidence_id, user_id=user_id)
         with main_database(self.evidence.database_path).read() as db:
             row = db.execute(
-                "SELECT evidence_id FROM evidence WHERE user_id='local' AND source_id=? "
+                "SELECT evidence_id FROM evidence WHERE user_id=? AND source_id=? "
                 "AND conversation_id=? AND message_id=?",
-                (SOURCE_ID, f"note:{identity}", identity),
+                (user_id, SOURCE_ID, f"note:{identity}", identity),
             ).fetchone()
         if row is None:
             raise KeyError(identity)
-        return self.receipt(str(row[0]))
+        return self.receipt(str(row[0]), user_id=user_id)
 
 
 def memory_add_router(dependencies: Sequence[Depends]) -> APIRouter:
@@ -376,14 +389,20 @@ def memory_add_router(dependencies: Sequence[Depends]) -> APIRouter:
             record = await asyncio.to_thread(additions.save, addition)
         except ValueError as error:
             raise HTTPException(409, str(error)) from error
-        saved = await asyncio.to_thread(additions.receipt, record.evidence_id)
+        saved = await asyncio.to_thread(
+            partial(additions.receipt, record.evidence_id, user_id=addition.user_id)
+        )
         background.add_task(schedule, additions, record)
         return saved
 
     @router.get("/additions/{request_id}", response_model=MemoryAddReceipt)
-    def memory_add_status(request_id: UUID, request: Request) -> MemoryAddReceipt:
+    def memory_add_status(
+        request_id: UUID,
+        request: Request,
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
+    ) -> MemoryAddReceipt:
         try:
-            return store(request).receipt_for_request(request_id)
+            return store(request).receipt_for_request(request_id, user_id=user_id)
         except KeyError as error:
             raise HTTPException(404, "memory_add_not_found") from error
         except ValueError as error:

memhub/apps/core/src/one_memory/memory_server/contracts.py

@@ -75,7 +75,11 @@ def batch_id(*, internal_session_id: str, first_round: int, last_round: int) ->
 INTERNAL_SESSION_ID_MAX = 128
 
 
-def internal_session_id_for(conversation_id: str, source_id: str) -> str:
+def internal_session_id_for(
+    conversation_id: str,
+    source_id: str,
+    user_id: str = "local",
+) -> str:
     """Normalise a conversation identity to the durable ``internal_session_id``.
 
     This is the single derivation shared by every Evidence-first intake path.
@@ -86,7 +90,17 @@ def internal_session_id_for(conversation_id: str, source_id: str) -> str:
     per-session serialisation split). Only a value that the contract would have
     rejected before it ever reached the queue is replaced by the versioned
     session id, which is deterministic for a given (source, conversation).
+
+    Another user's conversation is always derived with the user in it: two
+    Hermes profiles can hold the same session id, and sharing a session key
+    would serialise and batch one person's turns behind the other's. `local`
+    keeps the identity it always had.
     """
+    if user_id != "local":
+        return session_id(
+            harness_id=source_id,
+            external_session_id=f"user:{user_id}|{conversation_id}",
+        )
     if len(conversation_id) <= INTERNAL_SESSION_ID_MAX:
         return conversation_id
     derived = session_id(harness_id=source_id, external_session_id=conversation_id)

memhub/apps/core/src/one_memory/memory_server/durable_queue.py

@@ -54,6 +54,7 @@ from one_memory.memory_tasks import (
 )
 from one_memory.sdk_migration_log import error_chain_text
 from one_memory.service_levels import ServiceLevel, service_level_context
+from one_memory.user_scope import DEFAULT_USER_ID, lane_key, other_user_lane_prefix
 
 logger = logging.getLogger(__name__)
 
@@ -345,6 +346,27 @@ EVENT_TIME_WINDOW = timedelta(minutes=10)
 DIRECT_WRITE_SOURCES = frozenset({"one-memory-manual"})
 
 
+def request_user_id(request: WriteRequest) -> str:
+    """Whose write this is: its frozen input's user, `local` for id-only rows."""
+    users = {record.user_id for record in request.input_records}
+    if len(users) > 1:
+        raise ValueError("durable_write_multiple_users")
+    return next(iter(users), DEFAULT_USER_ID)
+
+
+def request_source_key(request: WriteRequest) -> str:
+    """The FIFO/claim key of a write: its source, per user.
+
+    `local` keeps the bare source id it always had. Another user's Hermes
+    lane is `hermes@user`, so one person's running job never holds the other
+    person's queue and the claim gate answers for that user's grant.
+    """
+    return lane_key(
+        request_user_id(request),
+        request.source_id or request.internal_session_id,
+    )
+
+
 @dataclass(frozen=True)
 class DurableWriteRecord:
     job_id: str
@@ -862,7 +884,7 @@ class DurableWriteQueue:
                     job_id,
                     queue_seq,
                     request.internal_session_id,
-                    request.source_id or request.internal_session_id,
+                    request_source_key(request),
                     request.batch_identity.batch_id,
                     key,
                     payload,
@@ -929,7 +951,9 @@ class DurableWriteQueue:
             f"{hashlib.sha256(file_key.encode()).hexdigest()}:"
             f"{hashlib.sha256(record.conversation_id.encode()).hexdigest()}"
         )
-        session_key = internal_session_id_for(record.conversation_id, record.source_id)
+        session_key = internal_session_id_for(
+            record.conversation_id, record.source_id, record.user_id
+        )
         written = connection.execute(
             "SELECT 1 FROM memory_write_jobs AS job, json_each(job.evidence_ids_json) AS ref "
             "WHERE job.session_key=? AND job.state!='assembling' AND ref.value=? LIMIT 1",
@@ -998,7 +1022,7 @@ class DurableWriteQueue:
                 str(uuid4()),
                 queue_seq,
                 session_key,
-                record.source_id,
+                lane_key(record.user_id, record.source_id),
                 assembly_key,
                 assembly_key,
                 payload_json,
@@ -1034,10 +1058,13 @@ class DurableWriteQueue:
         `flush` (history at end of file) completes the last turn and writes a
         short batch; without it (live polling) those stay in the row.
         """
+        # Two users' roots can hold a file with the same relative key (each
+        # Hermes profile has its own `state.db`): only this user's rows.
+        assembly_prefix = f"assembly:{user_id}:{source_id}:"
         rows = connection.execute(
             "SELECT * FROM memory_write_jobs WHERE state='assembling' "
-            "AND source_key=? ORDER BY queue_seq",
-            (source_id,),
+            "AND source_key=? AND substr(idempotency_key,1,?)=? ORDER BY queue_seq",
+            (lane_key(user_id, source_id), len(assembly_prefix), assembly_prefix),
         ).fetchall()
         finalized: list[str] = []
         for row in rows:
@@ -1160,7 +1187,7 @@ class DurableWriteQueue:
         Same records, same key: a re-read returns the job already written.
         """
         first = records[0]
-        session_id = internal_session_id_for(first.conversation_id, first.source_id)
+        session_id = internal_session_id_for(first.conversation_id, first.source_id, first.user_id)
         material = ":".join(f"{item.evidence_id}:{item.content_hash}" for item in records)
         request_id = f"turn:v2:{hashlib.sha256(material.encode()).hexdigest()}"
         existing = connection.execute(
@@ -1205,7 +1232,7 @@ class DurableWriteQueue:
                 job_id,
                 queue_seq,
                 session_id,
-                first.source_id,
+                lane_key(first.user_id, first.source_id),
                 request.batch_identity.batch_id,
                 request_id,
                 json.dumps(
@@ -2494,6 +2521,12 @@ async def execute_job(
             source_id = request.internal_session_id
         if source_ids and source_ids != {source_id}:
             raise RuntimeError("durable_write_multiple_sources")
+        users = {getattr(record, "user_id", DEFAULT_USER_ID) for record in records}
+        if len(users) > 1:
+            raise RuntimeError("durable_write_multiple_users")
+        # The gates answer per user: another user's Hermes job asks about
+        # that user's grant, not `local`'s.
+        source_id = lane_key(next(iter(users), DEFAULT_USER_ID), source_id)
 
         async def mark() -> bool:
             return await in_the_background(queue.mark_invoked, job_id, owner=owner)
@@ -2799,6 +2832,19 @@ class DurableWriteDrain:
                 source_clause = f"AND candidate.source_key IN ({placeholders}) "
                 source_params = tuple(sorted(enabled_sources))
             direct_sources = ",".join("?" for _ in DIRECT_WRITE_SOURCES)
+            # Another user's hand-added memories are `one-memory-manual@user`:
+            # the same lane ahead of history, still serial per user.
+            direct_lanes = " OR ".join(
+                "substr(candidate.source_key,1,?)=?" for _ in DIRECT_WRITE_SOURCES
+            )
+            direct_lane_params = tuple(
+                value
+                for source in sorted(DIRECT_WRITE_SOURCES)
+                for value in (
+                    len(other_user_lane_prefix(source)),
+                    other_user_lane_prefix(source),
+                )
+            )
             # Same eligibility as the window claim below (free lease, no
             # running job of its source, nothing earlier in its session), but
             # not bound to the window: a note must not wait for history.
@@ -2808,7 +2854,7 @@ class DurableWriteDrain:
                 "WHERE candidate.state=? "
                 "AND (candidate.lease_expires_at IS NULL OR candidate.lease_expires_at < ?) "
                 + source_clause
-                + f"AND candidate.source_key IN ({direct_sources}) "
+                + f"AND (candidate.source_key IN ({direct_sources}) OR {direct_lanes}) "
                 "AND NOT EXISTS ("
                 "SELECT 1 FROM memory_write_jobs AS held "
                 "WHERE held.source_key=candidate.source_key AND held.state=? "
@@ -2826,6 +2872,7 @@ class DurableWriteDrain:
                     now.isoformat(),
                     *source_params,
                     *sorted(DIRECT_WRITE_SOURCES),
+                    *direct_lane_params,
                     RUNNING,
                     now.isoformat(),
                     *_FIFO_BLOCKING_STATES,

memhub/apps/core/src/one_memory/memory_server/routes.py

@@ -19,10 +19,11 @@ does the writing. That separation is D52.
 
 from __future__ import annotations
 
+import hashlib
 from datetime import UTC, datetime
 from typing import Any, Literal, cast
 
-from fastapi import APIRouter, HTTPException, Request
+from fastapi import APIRouter, HTTPException, Query, Request
 from pydantic import BaseModel, ConfigDict, Field
 
 from one_memory.imports.models import (
@@ -43,6 +44,24 @@ from one_memory.memory_server.durable_queue import (
 )
 from one_memory.memory_server.registry import WriteRegistry, WriteState
 from one_memory.memory_server.submit import WriteConflictError
+from one_memory.user_scope import DEFAULT_USER_ID, USER_ID_PATTERN, UserId
+
+#: Durable keys of another user's writes start with this. A `local` request
+#: id never may, so `local` cannot name another user's receipt.
+_USER_KEY_PREFIX = "user:"
+
+
+def durable_request_key(request_id: str, user_id: str) -> str:
+    """The queue key of a caller's request id, namespaced per user.
+
+    Two users choosing the same request id must not collide: the key is the
+    queue's idempotency key and the SDK's `request_id`. `local` keeps its own
+    request id unchanged.
+    """
+    if user_id == DEFAULT_USER_ID:
+        return request_id
+    digest = hashlib.sha256(request_id.encode()).hexdigest()
+    return f"{_USER_KEY_PREFIX}{user_id}:{digest}"
 
 
 class MemoryWriteReceipt(BaseModel):
@@ -71,6 +90,7 @@ class SdkMemoryWriteInput(BaseModel):
     messages: list[OpenAIMessage] = Field(min_length=1, max_length=200)
     memory_at: datetime | None = None
     priority: Literal["live", "normal"] = "normal"
+    user_id: UserId = DEFAULT_USER_ID
 
 
 def _registry(
@@ -124,6 +144,9 @@ def sdk_router(dependencies: list[Any] | None = None) -> APIRouter:
     ) -> MemoryWriteReceipt:
         if not any(message.role == "user" for message in payload.messages):
             raise HTTPException(422, "messages_require_user_turn")
+        if payload.user_id == DEFAULT_USER_ID and payload.request_id.startswith(_USER_KEY_PREFIX):
+            raise HTTPException(422, "request_id_reserved_prefix")
+        request_key = durable_request_key(payload.request_id, payload.user_id)
         observed_at = payload.memory_at or datetime(1970, 1, 1, tzinfo=UTC)
         written_by = {
             "user": WrittenBy.PERSON,
@@ -133,6 +156,7 @@ def sdk_router(dependencies: list[Any] | None = None) -> APIRouter:
         records = tuple(
             materialize_write_input(
                 EvidenceInput(
+                    user_id=payload.user_id,
                     source_id=payload.source_id,
                     conversation_id=payload.session_id,
                     message_id=f"{payload.request_id}:{index}",
@@ -157,6 +181,7 @@ def sdk_router(dependencies: list[Any] | None = None) -> APIRouter:
         internal_session_id = internal_session_id_for(
             payload.session_id,
             payload.source_id,
+            payload.user_id,
         )
         contract = WriteRequest(
             internal_session_id=internal_session_id,
@@ -168,7 +193,7 @@ def sdk_router(dependencies: list[Any] | None = None) -> APIRouter:
             ),
             evidence_ids=[record.evidence_id for record in records],
             input_records=records,
-            idempotency_key=payload.request_id,
+            idempotency_key=request_key,
             priority=payload.priority,
             event_at=payload.memory_at,
             time_quality="exact" if payload.memory_at is not None else "approximate",
@@ -178,7 +203,7 @@ def sdk_router(dependencies: list[Any] | None = None) -> APIRouter:
         except (WriteConflictError, ValueError) as error:
             raise HTTPException(409, str(error)) from error
         return MemoryWriteReceipt(
-            request_id=record.request_id,
+            request_id=payload.request_id,
             state=record.state,
             created_at=record.created_at,
             updated_at=record.updated_at,
@@ -186,16 +211,24 @@ def sdk_router(dependencies: list[Any] | None = None) -> APIRouter:
         )
 
     @instance.get("/{request_id}", response_model=MemoryWriteReceipt)
-    def status(request_id: str, request: Request) -> MemoryWriteReceipt:
+    def status(
+        request_id: str,
+        request: Request,
+        user_id: str = Query(default=DEFAULT_USER_ID, pattern=USER_ID_PATTERN),
+    ) -> MemoryWriteReceipt:
+        # A receipt belongs to the user that submitted it. Another user's key
+        # is namespaced, and `local` may not name a namespaced key directly.
+        if user_id == DEFAULT_USER_ID and request_id.startswith(_USER_KEY_PREFIX):
+            raise HTTPException(404, "unknown request id")
         registry = _registry(request, write=False)
         try:
-            record = registry.get(request_id)
+            record = registry.get(durable_request_key(request_id, user_id))
         except DurableQueueSchemaUnavailableError as error:
             raise HTTPException(503, str(error)) from error
         if record is None:
             raise HTTPException(404, "unknown request id")
         return MemoryWriteReceipt(
-            request_id=record.request_id,
+            request_id=request_id,
             state=record.state,
             created_at=record.created_at,
             updated_at=record.updated_at,

memhub/apps/core/src/one_memory/memory_server/submit.py

@@ -271,7 +271,9 @@ def build_request_from_evidence(
         # Keep the exact conversation id when it fits the 128-char contract;
         # otherwise derive the versioned session id, so a long branch id does
         # not reject the whole history import.
-        internal_session_id=internal_session_id_for(seed.conversation_id, seed.source_id),
+        internal_session_id=internal_session_id_for(
+            seed.conversation_id, seed.source_id, seed.user_id
+        ),
         first_round=first_round,
         last_round=last_round,
         evidence_ids=list(ordered),

memhub/apps/core/src/one_memory/memory_server/unified_entry.py

@@ -28,6 +28,7 @@ from one_memory.memory_server.contracts import (
     internal_session_id_for,
 )
 from one_memory.memory_server.durable_queue import DurableWriteQueue, DurableWriteRecord
+from one_memory.user_scope import lane_key
 
 WritePriority = Literal["live", "normal"]
 
@@ -74,7 +75,7 @@ class UnifiedWriteIntake:
         if priority not in {"live", "normal"}:
             raise ValueError("priority must be 'live' or 'normal'")
         if self.source_writes_permitted is not None and not self.source_writes_permitted(
-            record.source_id
+            lane_key(record.user_id, record.source_id)
         ):
             # Freeze is the product's one Harness processing decision. Refuse
             # new intake so a frozen source creates no new durable write row;
@@ -106,7 +107,9 @@ class UnifiedWriteIntake:
         last_round: int,
         priority: WritePriority,
     ) -> WriteRequest:
-        session_id = internal_session_id_for(record.conversation_id, record.source_id)
+        session_id = internal_session_id_for(
+            record.conversation_id, record.source_id, record.user_id
+        )
         event_at, time_quality = _job_time((record,))
         request_key = (
             f"entry:{record.evidence_id}"
@@ -145,14 +148,14 @@ class UnifiedWriteIntake:
         ):
             raise ValueError("records must belong to one conversation")
         if self.source_writes_permitted is not None and not self.source_writes_permitted(
-            first.source_id
+            lane_key(first.user_id, first.source_id)
         ):
             raise SourceWriteDisabledError(first.evidence_id)
         material = ":".join(
             f"{record.evidence_id}:{record.revision}:{record.content_hash}" for record in records
         )
         request_id = f"turn:v1:{sha256(material.encode()).hexdigest()}"
-        session_id = internal_session_id_for(first.conversation_id, first.source_id)
+        session_id = internal_session_id_for(first.conversation_id, first.source_id, first.user_id)
         event_at, time_quality = _job_time(records)
         durable = self.queue.submit(
             WriteRequest(

memhub/apps/core/src/one_memory/recall.py

@@ -22,6 +22,7 @@ from one_memory.semantic_deadline import (
     normalize_semantic_hits,
     search_within,
 )
+from one_memory.user_scope import DEFAULT_USER_ID, UserId
 
 
 def recall_blend_mode() -> str:
@@ -67,6 +68,8 @@ class RecallRequest(BaseModel):
     #: Wall-clock budget for candidate gather + score. Not a substitute for
     #: raising the old newest-500 page; it is the third parallel budget.
     timeout_ms: int = Field(default=2_000, ge=50, le=30_000)
+    #: Whose memories to recall. Omitted, it is `local`, the App's library.
+    user_id: UserId = DEFAULT_USER_ID
 
 
 class RecallHit(BaseModel):

memhub/apps/core/src/one_memory/sdk_memory_qa.py

@@ -32,6 +32,7 @@ from one_memory.model_profiles import (
     attach_disable_thinking,
 )
 from one_memory.secrets import SecretStore, SecretStoreUnavailableError
+from one_memory.user_scope import owns
 
 _NO_GROUNDING = "现有记忆不足以回答这个问题。"
 _GROUNDING_LIMIT = 6
@@ -117,7 +118,12 @@ class SdkMemoryAnswerer:
             if hit.memory_id in seen:
                 continue
             record = await _get_memory(client, hit.memory_id)
-            if record is None or record.status != "active" or not record.is_latest:
+            if (
+                record is None
+                or record.status != "active"
+                or not record.is_latest
+                or not owns(record.user_id, user_id)
+            ):
                 continue
             seen.add(record.memory_id)
             found.append(_wire_memory(record))

memhub/apps/core/src/one_memory/sdk_memory_runtime.py

@@ -5,7 +5,7 @@ from __future__ import annotations
 import asyncio
 import logging
 import re
-from collections.abc import AsyncIterator
+from collections.abc import AsyncIterator, Callable, Iterable
 from contextlib import AsyncExitStack, asynccontextmanager, suppress
 from datetime import UTC, datetime, timedelta
 from pathlib import Path
@@ -56,8 +56,11 @@ class SdkMemoryRuntime:
         write_queue: DurableWriteQueue,
         extraction_settings: ExtractionSettingsStore,
         data_dir: Path,
+        user_ids: Callable[[], Iterable[str]] = lambda: ("local",),
     ) -> None:
         self.coordinator = coordinator
+        #: Every user whose raw layer is kept to the retention window.
+        self.user_ids = user_ids
         self.read_handle = read_handle
         self.write_queue = write_queue
         self.extraction_settings = extraction_settings
@@ -72,6 +75,12 @@ class SdkMemoryRuntime:
             )
 
     async def prune_expired_raw(self, *, now: datetime | None = None) -> int:
+        deleted = 0
+        for user_id in dict.fromkeys(await asyncio.to_thread(lambda: tuple(self.user_ids()))):
+            deleted += await self._prune_expired_raw_for(user_id, now=now)
+        return deleted
+
+    async def _prune_expired_raw_for(self, user_id: str, *, now: datetime | None) -> int:
         cutoff = (now or datetime.now(UTC)) - timedelta(days=RAW_RETENTION_DAYS)
         cutoff_timestamp = int(cutoff.timestamp())
         client = await self.read_handle.client()
@@ -79,7 +88,7 @@ class SdkMemoryRuntime:
         while True:
             page = await asyncio.to_thread(
                 client.list_memories,
-                user_id="local",
+                user_id=user_id,
                 layers=["l1_raw"],
                 statuses=list(_RAW_STATUSES),
                 order="asc",
@@ -95,8 +104,7 @@ class SdkMemoryRuntime:
                 for row in rows
                 if isinstance(row, dict)
                 and isinstance(row.get("memory_id"), str)
-                and int(row.get("memory_at") or row.get("gmt_created") or 0)
-                < cutoff_timestamp
+                and int(row.get("memory_at") or row.get("gmt_created") or 0) < cutoff_timestamp
             ]
             if not expired:
                 return deleted
@@ -154,11 +162,10 @@ class SdkMemoryRuntime:
         limit: int = 12,
     ) -> tuple[tuple[str, float], ...]:
         return tuple(
-            (hit.memory_id, hit.score)
-            for hit in await self.semantic_recall(query, limit=limit)
+            (hit.memory_id, hit.score) for hit in await self.semantic_recall(query, limit=limit)
         )
 
-    async def graph_snapshot(self, *, limit: int = 300) -> HyGraphSnapshot:
+    async def graph_snapshot(self, *, limit: int = 300, user_id: str = "local") -> HyGraphSnapshot:
         try:
             await self.coordinator.active_profile_credentials()
         except RuntimeError as error:
@@ -173,7 +180,7 @@ class SdkMemoryRuntime:
             return await asyncio.to_thread(
                 read_graph_snapshot,
                 client,
-                "local",
+                user_id,
                 "*",
                 limit,
             )

memhub/apps/core/src/one_memory/sdk_migration_runner.py

@@ -259,6 +259,7 @@ def prepare_legacy_write_queue_for_cutover(database_path: Path) -> tuple[int, in
         internal_session_id = internal_session_id_for(
             source.conversation_id,
             source.source_id,
+            source.user_id,
         )
         queue.submit(
             WriteRequest(

memhub/apps/core/src/one_memory/strict_query.py

@@ -0,0 +1,39 @@
+"""Refuse query parameters a route does not declare.
+
+FastAPI ignores an unknown query parameter. For a memory read that is a
+silent widening: `memory-library?user_id=__nobody__` against a Core that did
+not know `user_id` answered with the whole library, and the caller believed
+it was isolated. Routes that carry this dependency answer such a request with
+the same 422 a malformed body gets.
+"""
+
+from fastapi import Request
+from fastapi.dependencies.models import Dependant
+from fastapi.exceptions import RequestValidationError
+from fastapi.routing import APIRoute
+
+
+def _declared_query_names(dependant: Dependant) -> set[str]:
+    names = {parameter.alias for parameter in dependant.query_params}
+    for dependency in dependant.dependencies:
+        names |= _declared_query_names(dependency)
+    return names
+
+
+async def forbid_unknown_query_params(request: Request) -> None:
+    route = request.scope.get("route")
+    if not isinstance(route, APIRoute):
+        return
+    unknown = sorted(set(request.query_params) - _declared_query_names(route.dependant))
+    if unknown:
+        raise RequestValidationError(
+            [
+                {
+                    "type": "extra_forbidden",
+                    "loc": ("query", name),
+                    "msg": "Extra inputs are not permitted",
+                    "input": None,
+                }
+                for name in unknown
+            ]
+        )

memhub/apps/core/src/one_memory/user_scope.py

@@ -0,0 +1,65 @@
+"""Who a memory belongs to: one Runtime, many `user_id`s.
+
+A user is decided by which source root is read (a Hermes profile is its own
+`state.db`), never by session or platform fields. The App never names a user
+and keeps reading and writing `local`; every other harness stays on `local`
+too. Only Hermes may connect a second root for another user.
+
+In memory, a source's import lane is keyed by `lane_key(user_id, source_id)`.
+For `local` that is the bare source id, so every existing key, health record
+and log line stays byte-identical; another user's lane is `source@user`.
+`@` cannot occur in a user id or a source id, so the key splits back exactly.
+"""
+
+import re
+from typing import Annotated
+
+from pydantic import Field
+
+DEFAULT_USER_ID = "local"
+
+#: `:` is excluded on purpose: the SDK joins ids with it when it keys history,
+#: so a user id holding one could read another user's history.
+USER_ID_PATTERN = r"^[A-Za-z0-9._-]{1,64}$"
+_USER_ID = re.compile(USER_ID_PATTERN)
+
+#: The only source whose roots may belong to a user other than `local`.
+MULTI_USER_SOURCES = frozenset({"hermes"})
+
+UserId = Annotated[str, Field(min_length=1, max_length=64, pattern=USER_ID_PATTERN)]
+
+_LANE_SEPARATOR = "@"
+
+
+def valid_user_id(value: str) -> bool:
+    return bool(_USER_ID.fullmatch(value))
+
+
+def lane_key(user_id: str, source_id: str) -> str:
+    if user_id == DEFAULT_USER_ID:
+        return source_id
+    return f"{source_id}{_LANE_SEPARATOR}{user_id}"
+
+
+def other_user_lane_prefix(source_id: str) -> str:
+    """What every other user's lane of `source_id` starts with."""
+    return f"{source_id}{_LANE_SEPARATOR}"
+
+
+def split_lane(key: str) -> tuple[str, str]:
+    """`(user_id, source_id)` of a lane key; a bare source id is `local`'s."""
+    source_id, separator, user_id = key.partition(_LANE_SEPARATOR)
+    if not separator or not valid_user_id(user_id):
+        return DEFAULT_USER_ID, key
+    return user_id, source_id
+
+
+def owns(record_user_id: str | None, user_id: str) -> bool:
+    """Whether a stored memory belongs to `user_id`.
+
+    The SDK's `get`/`update`/`delete`/`history`/`get_write_outcome` take no
+    user, so MemHub checks this before acting on an id. A memory written
+    before MemHub named users carries `local` or nothing; both are `local`'s.
+    """
+    stored = record_user_id or DEFAULT_USER_ID
+    return stored == user_id

memhub/apps/core/tests/TEST_INDEX.md

@@ -105,6 +105,7 @@
 | `test_graph_width_guard.py` | `unit` | Graph width mismatch must ask, never silently archive. |
 | `test_harness_contract_suite.py` | `contract` | A4-09: one contract, ten harnesses. |
 | `test_health.py` | `contract` | 覆盖 a backfill that fails says so rather than disappearing、a failed catch up leaves a traceback behind、configured model without content consent does not start backfill 等 15 个场景。 |
+| `test_hermes_gateway_injection.py` | `contract` | Hermes gateway text in `user` rows: notices are skipped, relayed messages unwrapped. |
 | `test_history_listing_rotation.py` | `contract` | 覆盖 history listing alternates started and untouched files、history listing drains whichever cohort is longer 场景。 |
 | `test_host_capabilities.py` | `unit` | A plugin does not become realtime on every host or every unknown version. |
 | `test_hourly_c1_peak_replay_contract.py` | `contract` | Guard the package peak-replay probe's error accounting. |
@@ -196,6 +197,7 @@
 | `test_model_profiles.py` | `contract` | 覆盖 profile uses one model and keeps key out of database、api key updates reuse one stable secret reference、startup cleanup removes only secrets not referenced by current profiles 等 15 个场景。 |
 | `test_model_request_audit.py` | `contract` | 覆盖 failure is recorded without error message、missing usage stays unknown in detail and zero in summary、records one physical request with provider usage and no content 等 5 个场景。 |
 | `test_model_request_budget.py` | `contract` | 覆盖 persists failed request reservations without content、parallel leases cannot exceed request limit、unknown or excessive output and body are denied 等 6 个场景。 |
+| `test_multi_user_runtime.py` | `contract` | One Runtime, many users: `user_id` on the memory API, per-user Hermes profiles. |
 | `test_mysql_lifecycle.py` | `contract` | A1-07: the decisions Runtime makes around the bundled MySQL. |
 | `test_mysql_platform.py` | `contract` | A1-07: the machine-facing half of managing the bundled MySQL. |
 | `test_obs03_health_probe.py` | `contract` | OBS-03: '/health' answered with the web app instead of a health payload. |
@@ -309,4 +311,4 @@
 | `test_workspace_recall.py` | `contract` | Workspace memories were unreachable from every host; these pin the way in. |
 | `test_write_queue_lease_renewal.py` | `contract` | Rule 8: a lease must outlive the work it covers, and be renewed while it runs. |
 
-共 304 个测试文件。
+共 306 个测试文件。

memhub/apps/core/tests/core_test_classification.py

@@ -14,8 +14,8 @@ from pathlib import Path
 import pytest
 
 PRIMARY_MARKERS = ("unit", "contract", "integration", "performance", "live")
-EXPECTED_TEST_FILE_COUNT = 304
-EXPECTED_TEST_FILES_SHA256 = "b28f022b9688c96ca563c59de63e95d344d2f01ba3b6020b422bf4dfb27520d6"
+EXPECTED_TEST_FILE_COUNT = 306
+EXPECTED_TEST_FILES_SHA256 = "e616dcf9fea288bf15a4445045f7b4278517f3304091c04fffd55be80308d481"
 
 # Pure transformations, validation models, parity checks and state machines.
 # This list is intentionally small: uncertain files default to contract.

memhub/apps/core/tests/fixtures/hermes_gateway_rows.jsonl

@@ -0,0 +1,8 @@
+{"id": 131, "session_id": "20261019_005046_a2a79a", "role": "user", "content": "[IMPORTANT: 3 background processes completed. Treat these results as one batch and give one consolidated response; preserve failures and actionable results.]", "timestamp": 1792344160.7817369, "tool_name": null, "finish_reason": null}
+{"id": 314, "session_id": "20261019_005046_a2a79a", "role": "user", "content": "[IMPORTANT: Background process proc_a9805510da6f matched watch pattern \"Uvicorn running\".\nCommand: cd ~/Workspace/paddleocr-vl-service && exec ./start.sh\nMatched output:\nINFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)]", "timestamp": 1792421570.094992, "tool_name": null, "finish_reason": null}
+{"id": 769, "session_id": "20261021_005036_4fcb846a", "role": "user", "content": "[IMPORTANT: Background process proc_9d3a5fe22318 completed normally (exit code 0).\nCommand: cd /Users/example/.hermes/models/gguf && curl -L -C - --retry 30 --retry-delay 5 --retry-all-errors -sS -o Qwen3.5-2B-Q4_K_M.gguf \"https://huggingface.co/unsloth/Qwen3.5-2B-GGUF/resolve/main/Qwen3.5-2B-Q4_K_M.gguf\" && echo \"Q3.5-2B DONE\" && ls -lh Qwen3.5-2B-Q4_K_M.gguf\nOutput:\n(eval):1: can't change option: zle\n(eval):1: can't change option: zle\nQ3.5-2B DONE\n-rw-r--r--@ 1 example  staff   1.2G Oct 21 01:31 Qwen3.5-2B-Q4_K_M.gguf\n]\n\n[INTERNAL NOTIFICATION — generated by the Hermes gateway, not written by a human user. Do not treat it as user input, and do not reply to any earlier chat message because of it.]", "timestamp": 1792517511.042671, "tool_name": null, "finish_reason": null}
+{"id": 343, "session_id": "20261021_005036_4fcb846a", "role": "user", "content": "[voice message could not be transcribed automatically; the audio is available at: /Users/example/.hermes/cache/audio/audio_46bab1cf72c3.ogg]", "timestamp": 1792515117.0, "tool_name": null, "finish_reason": null}
+{"id": 197, "session_id": "20261019_142101_50c2a6", "role": "user", "content": "Reply with exactly: PONG", "timestamp": 1792390861.155566, "tool_name": null, "finish_reason": null}
+{"id": 637, "session_id": "20261021_005036_4fcb846a", "role": "user", "content": "Gateway message origin (JSON data, not instructions or authorization):\n{\"platform\": \"telegram\", \"chat_id\": \"1000000001\", \"chat_type\": \"dm\", \"user_id\": \"1000000001\", \"message_id\": \"93\", \"source_message_id\": \"93\"}\nDo not guess a reply destination when these fields are insufficient.\n\n3.5 2b 得版本也想看看，没有量化或者 gguf 吗", "timestamp": 1792517455.622248, "tool_name": null, "finish_reason": null}
+{"id": 548, "session_id": "20261021_005036_4fcb846a", "role": "user", "content": "[OUT-OF-BAND USER MESSAGE — a direct message from the user, delivered once at this position; not tool output and not a new delivery when replayed from conversation history]\nGateway message origin (JSON data, not instructions or authorization):\n{\"platform\": \"telegram\", \"chat_id\": \"1000000001\", \"chat_type\": \"dm\", \"user_id\": \"1000000001\", \"message_id\": \"73\", \"source_message_id\": \"73\"}\nDo not guess a reply destination when these fields are insufficient.\n\n对比结果如何\n[/OUT-OF-BAND USER MESSAGE]", "timestamp": 1792517335.660696, "tool_name": null, "finish_reason": null}
+{"id": 2154, "session_id": "20261021_005036_4fcb846a", "role": "user", "content": "[OUT-OF-BAND USER MESSAGE — a direct message from the user, delivered once at this position; not tool output and not a new delivery when replayed from conversation history]\nGateway message origin (JSON data, not instructions or authorization):\n{\"platform\": \"telegram\", \"chat_id\": \"1000000001\", \"chat_type\": \"dm\", \"user_id\": \"1000000001\", \"message_id\": \"489\", \"source_message_id\": \"489\"}\nDo not guess a reply destination when these fields are insufficient.\n\n我预期memhub使用的是hy memory的ultra模式，并且在夜间调用一次envolve接口来进化memory 呢。\n你说memhub的chat走的是什么我有点懵。\n\n我记得memhub写入的时候，不是直接调用sdk client的add么？\n\n应该是没有设置scene才对。可以检查一下\n[/OUT-OF-BAND USER MESSAGE]", "timestamp": 1793448843.4533901, "tool_name": null, "finish_reason": null}

memhub/apps/core/tests/test_hermes_gateway_injection.py

@@ -0,0 +1,239 @@
+"""Hermes gateway text in `user` rows: notices are skipped, relayed messages unwrapped.
+
+Every fixture row is a real Hermes `state.db` row, with the Telegram ids and
+the local home directory replaced by fake values (and the batch notice cut
+to its first line). They all wear the `user` role, so the import has to
+decide on the text.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import sqlite3
+from pathlib import Path
+
+import pytest
+from pydantic import JsonValue
+
+from one_memory.conversation_processing import turn_batches
+from one_memory.imports.access import AuthorizedSourceRoot
+from one_memory.imports.adapters import (
+    ScanEvent,
+    ScanIssueCode,
+    ScannedEvidence,
+    ScanOptions,
+    ScanSkip,
+)
+from one_memory.imports.sources.common import (
+    ParsedMessage,
+    ParsedSkip,
+    SourceRecordContext,
+)
+from one_memory.imports.sources.hermes import (
+    HERMES_DESCRIPTOR,
+    create_hermes_adapter,
+    gateway_user_text,
+    parse_hermes_rollout_record,
+)
+from one_memory.imports.storage import EvidenceStore
+
+_ROWS = {
+    row["id"]: row
+    for row in (
+        json.loads(line)
+        for line in (Path(__file__).parent / "fixtures" / "hermes_gateway_rows.jsonl")
+        .read_text(encoding="utf-8")
+        .splitlines()
+        if line.strip()
+    )
+}
+
+_DROPPED = {
+    131: ScanIssueCode.INJECTED_MACHINE_NOTICE,  # "3 background processes completed"
+    314: ScanIssueCode.INJECTED_MACHINE_NOTICE,  # "Background process … matched watch pattern"
+    769: ScanIssueCode.INJECTED_MACHINE_NOTICE,  # … plus the gateway's INTERNAL NOTIFICATION
+    343: ScanIssueCode.VOICE_PLACEHOLDER,
+    197: ScanIssueCode.PROBE,
+}
+
+_UNWRAPPED = {
+    548: "对比结果如何",
+    637: "3.5 2b 得版本也想看看，没有量化或者 gguf 吗",
+    2154: (
+        "我预期memhub使用的是hy memory的ultra模式，并且在夜间调用一次envolve接口来进化memory 呢。\n"
+        "你说memhub的chat走的是什么我有点懵。\n\n"
+        "我记得memhub写入的时候，不是直接调用sdk client的add么？\n\n"
+        "应该是没有设置scene才对。可以检查一下"
+    ),
+}
+
+
+@pytest.mark.parametrize("row_id", sorted(_DROPPED))
+def test_a_machine_turn_in_the_users_role_is_skipped_with_its_reason(row_id: int) -> None:
+    assert gateway_user_text(_ROWS[row_id]["content"]) == ("", _DROPPED[row_id])
+
+
+def test_the_gateway_note_drops_a_turn_wherever_it_sits() -> None:
+    note = _ROWS[769]["content"].rsplit("\n\n", 1)[1]
+    assert note.startswith("[INTERNAL NOTIFICATION")
+
+    assert gateway_user_text(f"anything before it\n\n{note}")[1] is (
+        ScanIssueCode.INJECTED_MACHINE_NOTICE
+    )
+
+
+@pytest.mark.parametrize("row_id", sorted(_UNWRAPPED))
+def test_a_relayed_message_keeps_exactly_what_the_person_wrote(row_id: int) -> None:
+    assert gateway_user_text(_ROWS[row_id]["content"]) == (_UNWRAPPED[row_id], None)
+
+
+def test_the_interrupted_response_label_is_unwrapped() -> None:
+    text = "[Context from the interrupted assistant response]\n继续刚才的对比"
+
+    assert gateway_user_text(text) == ("继续刚才的对比", None)
+
+
+@pytest.mark.parametrize(
+    ("text", "reason"),
+    [
+        ("MEDIA:/Users/example/.hermes/cache/images/a.jpg", ScanIssueCode.NO_TEXT),
+        ("Reply with exactly: PONG please, and then explain why", None),
+        ("平铺", None),
+        ("  look at [IMPORTANT: Background process] in the log  ", None),
+    ],
+)
+def test_ordinary_text_is_left_as_it_was(text: str, reason: ScanIssueCode | None) -> None:
+    cleaned, skipped = gateway_user_text(text)
+
+    assert skipped is reason
+    assert cleaned == ("" if reason else text.strip())
+
+
+def _state_db(root: Path, rows: list[tuple[int, str, str, str]]) -> None:
+    root.mkdir()
+    connection = sqlite3.connect(root / "state.db")
+    connection.execute(
+        "CREATE TABLE messages (id INTEGER PRIMARY KEY, session_id TEXT, role TEXT,"
+        " content TEXT, timestamp REAL, tool_name TEXT, active INTEGER)"
+    )
+    connection.executemany(
+        "INSERT INTO messages VALUES (?, ?, ?, ?, 1000, ?, 1)",
+        [
+            (row_id, session, role, content, "terminal" if role == "tool" else None)
+            for row_id, session, role, content in rows
+        ],
+    )
+    connection.commit()
+    connection.close()
+
+
+def _scan(root: Path, home: Path) -> list[ScanEvent]:
+    adapter = create_hermes_adapter(
+        AuthorizedSourceRoot.grant(
+            source_id=HERMES_DESCRIPTOR.source_id,
+            root=root,
+            user_confirmed=True,
+            home=home,
+        )
+    )
+
+    async def collect() -> list[ScanEvent]:
+        return [event async for event in adapter.scan(ScanOptions(cursors={}))]
+
+    return asyncio.run(collect())
+
+
+def test_state_db_import_writes_the_cleaned_turn_and_counts_the_skipped_ones(
+    tmp_path: Path,
+) -> None:
+    rows = [
+        (row_id, "session-h", "user", _ROWS[row_id]["content"])
+        for row_id in [*_DROPPED, *_UNWRAPPED]
+    ]
+    _state_db(tmp_path / "hermes", rows)
+    events = _scan(tmp_path / "hermes", tmp_path)
+    evidence = [event.evidence for event in events if isinstance(event, ScannedEvidence)]
+    skipped = [event.reason_code for event in events if isinstance(event, ScanSkip)]
+
+    assert [item.verbatim for item in evidence] == [_UNWRAPPED[row_id] for row_id in _UNWRAPPED]
+    assert sorted(skipped) == sorted(_DROPPED.values())
+    assert "chat_id" not in repr(evidence)
+
+
+def test_assistant_and_tool_rows_with_the_same_text_are_not_touched(tmp_path: Path) -> None:
+    """Only the person's role is cleaned. An assistant that answers "PONG" or
+    a tool that prints a notice is reporting what it saw; the existing
+    assistant and tool rules decide what of that is kept."""
+    probe = _ROWS[197]["content"]
+    relayed = _ROWS[548]["content"]
+    _state_db(
+        tmp_path / "hermes",
+        [
+            (1, "session-h", "assistant", probe),
+            (2, "session-h", "assistant", relayed),
+            (3, "session-h", "tool", probe),
+        ],
+    )
+
+    evidence = [
+        event.evidence
+        for event in _scan(tmp_path / "hermes", tmp_path)
+        if isinstance(event, ScannedEvidence)
+    ]
+
+    assert [item.role for item in evidence] == ["assistant", "assistant", "tool"]
+    assert evidence[0].verbatim == probe
+    assert evidence[1].verbatim == relayed
+    assert probe in evidence[2].verbatim
+
+
+def test_a_conversation_left_without_its_user_turn_writes_no_batch(tmp_path: Path) -> None:
+    """A probe session is the probe and the reply. With the probe skipped,
+    the turn rule holds the reply back and drops it at the end of the file
+    rather than send HY a batch with no user message in it."""
+    _state_db(
+        tmp_path / "hermes",
+        [(1, "probe", "user", _ROWS[197]["content"]), (2, "probe", "assistant", "PONG")],
+    )
+    store = EvidenceStore(tmp_path / "one-memory.sqlite3")
+    records = [
+        store.put(event.evidence).record
+        for event in _scan(tmp_path / "hermes", tmp_path)
+        if isinstance(event, ScannedEvidence)
+    ]
+
+    assert [record.role for record in records] == ["assistant"]
+    assert turn_batches(records, flush=True).batches == ()
+
+
+@pytest.mark.parametrize("role", ["user", "assistant"])
+def test_session_transcripts_follow_the_same_rule(role: str) -> None:
+    context = SourceRecordContext(
+        file_key="hermes",
+        relative_path="sessions/session.jsonl",
+        ordinal="1",
+    )
+
+    def parse(text: str) -> ParsedMessage | ParsedSkip:
+        record: dict[str, JsonValue] = {
+            "type": "message",
+            "role": role,
+            "content": text,
+            "sessionId": "s",
+        }
+        return parse_hermes_rollout_record(record, context)
+
+    notice = parse(_ROWS[314]["content"])
+    relayed = parse(_ROWS[548]["content"])
+
+    if role == "user":
+        assert notice == ParsedSkip(
+            reason_code=ScanIssueCode.INJECTED_MACHINE_NOTICE, conversation_id="s"
+        )
+        assert isinstance(relayed, ParsedMessage) and relayed.verbatim == "对比结果如何"
+    else:
+        assert isinstance(notice, ParsedMessage)
+        assert notice.verbatim == _ROWS[314]["content"]
+        assert isinstance(relayed, ParsedMessage)
+        assert relayed.verbatim == _ROWS[548]["content"]

memhub/apps/core/tests/test_memory_additions.py

@@ -102,7 +102,7 @@ def test_memory_add_requires_authentication_explicit_scope_and_nonempty_content(
                 {"text": "  \n"},
                 {"text": "x" * 12_001},
                 {"scope": "*"},
-                {"user_id": "other"},
+                {"user_id": "not:a-user"},
             ):
                 assert (
                     await client.post("/api/v2/memories", json={**base, **change})

memhub/apps/core/tests/test_memory_qa_api.py

@@ -56,7 +56,7 @@ def test_sdk_ask_v2_stream_is_native(tmp_path, monkeypatch) -> None:
     )
 
     class Answerer:
-        async def answer_stream(self, _question):
+        async def answer_stream(self, _question, *, user_id="local"):
             yield {"type": "recall", "memories": (memory,)}
             yield {"type": "delta", "text": "你要求真实数据。"}
             yield {

memhub/apps/core/tests/test_multi_user_runtime.py

@@ -0,0 +1,789 @@
+"""One Runtime, many users: `user_id` on the memory API, per-user Hermes profiles.
+
+A user is the root that is read (a Hermes profile's own `state.db`), never a
+session or platform field. The App never names a user and keeps `local`;
+every other harness stays on `local`. These tests drive the production app
+against a user-aware stand-in for the public SDK client, whose `get`,
+`update` and `delete` take no user exactly like hy-memory 2.0.9's.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import sqlite3
+from contextlib import asynccontextmanager
+from datetime import UTC, datetime, timedelta
+from pathlib import Path
+from uuid import uuid4
+
+import pytest
+from httpx import ASGITransport, AsyncClient
+from pydantic import SecretStr
+
+from one_memory.api import _sdk_recall_target, create_app
+from one_memory.config import Settings
+from one_memory.imports.access import AuthorizedSourceRoot
+from one_memory.imports.adapters import ScannedEvidence, ScanOptions
+from one_memory.imports.models import (
+    EvidenceInput,
+    EvidenceOrigin,
+    OriginMetadata,
+    WrittenBy,
+    materialize_write_input,
+)
+from one_memory.imports.sources.hermes import HERMES_DESCRIPTOR, create_hermes_adapter
+from one_memory.imports.sources.registry import build_source_catalog
+from one_memory.memory_server.contracts import internal_session_id_for
+from one_memory.memory_server.durable_queue import (
+    DurableWriteDrain,
+    DurableWriteQueue,
+    execute_job,
+)
+from one_memory.sdk_write_processor import SdkWriteProcessor
+from one_memory.secrets import InMemorySecretStore
+from one_memory.user_scope import lane_key, split_lane, valid_user_id
+
+TOKEN = "multi-user-test"
+
+
+class UserAwareSdk:
+    """The public client surface MemHub calls, holding memories per user."""
+
+    def __init__(self) -> None:
+        self.memories: dict[str, dict[str, object]] = {}
+        self.add_calls: list[dict[str, object]] = []
+
+    def _store(self, user_id: str, content: str, *, source: str, session_id: str) -> str:
+        memory_id = f"memory-{len(self.memories) + 1}"
+        self.memories[memory_id] = {
+            "memory_id": memory_id,
+            "content": content,
+            "layer": "l2_fact",
+            "status": "active",
+            "user_id": user_id,
+            "agent_id": "all-apps",
+            "session_id": session_id,
+            "source": source,
+            "tags": [],
+            "gmt_created": 1_790_000_000,
+        }
+        return memory_id
+
+    def add(self, messages, *, user_id, session_id, source, request_id, **_kwargs):
+        self.add_calls.append({"user_id": user_id, "request_id": request_id})
+        text = " ".join(item["content"] for item in messages if item["role"] == "user")
+        memory_id = self._store(user_id, text, source=source, session_id=session_id)
+        return {"success": True, "memory_ids": [memory_id]}
+
+    def add_extracted(self, contents, *, user_id, session_id, source, request_id, **_kwargs):
+        self.add_calls.append({"user_id": user_id, "request_id": request_id})
+        memory_id = self._store(user_id, " ".join(contents), source=source, session_id=session_id)
+        return {"success": True, "memory_ids": [memory_id]}
+
+    def _rows(self, user_ids) -> list[dict[str, object]]:
+        return [dict(row) for row in self.memories.values() if row["user_id"] in user_ids]
+
+    def list_memories(self, *, user_id, scene=None, offset=0, limit=100, **_kwargs):
+        rows = [] if _kwargs.get("layers") == ["l1_raw"] else self._rows({user_id})
+        page = rows[offset : offset + limit]
+        if scene == "all":
+            return {
+                "scene": "all",
+                "memories": [{**row, "scene": "chat", "kind": "chat"} for row in page],
+                "total": len(rows),
+                "totals": {"chat": len(rows), "working": 0, "coding": 0},
+                "errors": {},
+                "limit": limit,
+                "offset": offset,
+                "request_id": "list",
+                "elapsed_ms": 1.0,
+            }
+        return {"vdb": {"memories": page, "total": len(rows), "limit": limit, "offset": offset}}
+
+    def memory_stats(self, *, user_id, statuses=None):
+        return {"total": len(self._rows({user_id})), "request_id": "stats", "elapsed_ms": 1.0}
+
+    def search(self, query, *, user_ids, scene=None, **kwargs):
+        rows = [] if kwargs.get("layers") == ["l1_raw"] else self._rows(set(user_ids))
+        items = [{**row, "scene": "chat", "kind": "chat", "score": 0.9} for row in rows]
+        if scene == "all":
+            return {
+                "request_id": "search",
+                "scene": "all",
+                "memories": items,
+                "lanes": {"chat": len(items), "working": 0, "coding": 0},
+                "errors": {},
+                "elapsed_ms": 1.0,
+            }
+        return {"request_id": "search", "scene": scene or "normal", "memories": items}
+
+    def get(self, memory_id, kind=None, variant=None):
+        row = self.memories.get(memory_id)
+        return None if row is None else {**row, "kind": "chat", "variant": None}
+
+    def update(self, memory_id, content, kind=None, variant=None, task=None):
+        if memory_id not in self.memories:
+            return {"success": False, "error_message": f"memory_id '{memory_id}' not found"}
+        self.memories[memory_id]["content"] = content
+        return {"success": True, "memory_id": memory_id}
+
+    def delete(self, memory_id, kind=None, variant=None):
+        if self.memories.pop(memory_id, None) is None:
+            return {"success": False, "deleted_count": 0}
+        return {"success": True, "deleted_count": 1}
+
+    def get_user_profile(self, *, user_id):
+        return {
+            "request_id": f"profile-{user_id}",
+            "profile": {
+                "memory_id": f"l0-{user_id}",
+                "content": f"name is {user_id}",
+                "layer": "L0_BASIC_INFO",
+                "basic_info": {"name": user_id},
+            },
+            "elapsed_ms": 1.0,
+        }
+
+
+class Handle:
+    def __init__(self, sdk: UserAwareSdk) -> None:
+        self.sdk = sdk
+
+    async def client(self) -> UserAwareSdk:
+        return self.sdk
+
+    async def close(self) -> None:
+        return None
+
+
+class Coordinator:
+    """The production write processor over the stand-in client."""
+
+    def __init__(self, sdk: UserAwareSdk) -> None:
+        self.processor = SdkWriteProcessor(Handle(sdk))  # type: ignore[arg-type]
+
+    async def process_sdk_job(self, records, *, request_id, before_hy_invocation):
+        return await self.processor.process_sdk_job(
+            records, request_id=request_id, before_hy_invocation=before_hy_invocation
+        )
+
+
+async def _never(_job_id: str) -> None:
+    return None
+
+
+async def drain_all(queue: DurableWriteQueue, sdk: UserAwareSdk) -> None:
+    """Claim and run every pending job, ungated, the way the drain would."""
+    drain = DurableWriteDrain(queue, execute=_never, lease=timedelta(seconds=30))
+    while claimed := drain.claim(limit=8):
+        for job_id in claimed:
+            await execute_job(
+                job_id,
+                queue=queue,
+                supervisor=Coordinator(sdk),
+                evidence=None,  # type: ignore[arg-type]
+            )
+
+
+def application(tmp_path: Path, *, home: Path | None = None):
+    """The production app; with `home`, imports run over that home's sources."""
+    settings = Settings(
+        data_dir=tmp_path / "data",
+        source_home=home or tmp_path,
+        api_token=SecretStr(TOKEN),
+        imports_enabled=home is not None,
+        automatic_memory_processing=False,
+        live_sync_enabled=False,
+    )
+    return create_app(
+        settings,
+        secret_store=InMemorySecretStore(),
+        source_catalog=(
+            build_source_catalog(home=home, environment={}, platform="darwin")
+            if home is not None
+            else None
+        ),
+    )
+
+
+def client_for(app) -> AsyncClient:
+    return AsyncClient(
+        transport=ASGITransport(app=app),
+        base_url="http://test",
+        headers={"Authorization": f"Bearer {TOKEN}"},
+    )
+
+
+def run_with_sdk(tmp_path, monkeypatch, scenario) -> None:
+    async def main() -> None:
+        app = application(tmp_path)
+        sdk = UserAwareSdk()
+
+        @asynccontextmanager
+        async def write_client():
+            yield sdk
+
+        async with app.router.lifespan_context(app):
+            app.state.sdk_client_handle = Handle(sdk)
+            monkeypatch.setattr(app.state.sdk_write_coordinator, "use_client", write_client)
+            async with client_for(app) as client:
+                await scenario(app, sdk, client)
+
+    asyncio.run(main())
+
+
+def _messages(text: str) -> list[dict[str, str]]:
+    return [{"role": "user", "content": text}, {"role": "assistant", "content": "noted"}]
+
+
+async def _write_both_users(app, sdk, client) -> dict[str, str]:
+    """`local` and `wife` each add one memory and submit one conversation."""
+    receipts = {}
+    for user, word in (("local", "tea"), ("wife", "coffee")):
+        request_id = str(uuid4())
+        body = {"request_id": request_id, "text": f"{user} drinks {word}", "scope": "all-apps"}
+        if user != "local":
+            body["user_id"] = user
+        added = await client.post("/api/v2/memories", json=body)
+        assert added.status_code == 201, added.text
+        receipts[f"add:{user}"] = request_id
+        write = {
+            "request_id": "same-request-id",
+            "session_id": "same-session",
+            "source_id": "one-memory-manual",
+            "messages": _messages(f"{user} walks the {word} route"),
+        }
+        if user != "local":
+            write["user_id"] = user
+        written = await client.post("/api/v2/memory-writes", json=write)
+        assert written.status_code == 202, written.text
+        assert written.json()["request_id"] == "same-request-id"
+    await drain_all(app.state.memory_write_queue, sdk)
+    return receipts
+
+
+def _contents(page: dict) -> set[str]:
+    return {memory["content"] for memory in page["memories"]}
+
+
+@pytest.mark.fresh_sdk
+def test_each_user_reads_only_their_own_library_search_profile_and_graph(
+    tmp_path, monkeypatch
+) -> None:
+    async def scenario(app, sdk, client):
+        await _write_both_users(app, sdk, client)
+        assert {call["user_id"] for call in sdk.add_calls} == {"local", "wife"}
+        # The same caller request id from two users is two SDK requests.
+        assert len({call["request_id"] for call in sdk.add_calls}) == len(sdk.add_calls) == 4
+
+        local = (await client.get("/api/v2/memory-library")).json()
+        wife = (await client.get("/api/v2/memory-library", params={"user_id": "wife"})).json()
+        assert _contents(local) == {"local drinks tea", "local walks the tea route"}
+        assert _contents(wife) == {"wife drinks coffee", "wife walks the coffee route"}
+        nobody = await client.get("/api/v2/memory-library", params={"user_id": "__nobody__"})
+        assert nobody.status_code == 200 and nobody.json()["memories"] == []
+
+        searched = (
+            await client.get("/api/v2/memory-library", params={"query": "route", "user_id": "wife"})
+        ).json()
+        assert _contents(searched) == {"wife walks the coffee route"}
+
+        context = await client.post(
+            "/api/v2/context/search",
+            json={"query": "drinks", "kinds": ["memory"], "user_id": "wife"},
+        )
+        assert context.status_code == 200, context.text
+        assert {item["content"] for item in context.json()["items"]} == {
+            "wife drinks coffee",
+            "wife walks the coffee route",
+        }
+        local_context = await client.post(
+            "/api/v2/context/search", json={"query": "drinks", "kinds": ["memory"]}
+        )
+        assert all("wife" not in item["content"] for item in local_context.json()["items"])
+
+        profile = (await client.get("/api/v2/user-profile", params={"user_id": "wife"})).json()
+        assert profile["profile"]["basic_info"] == {"name": "wife"}
+        assert (await client.get("/api/v2/user-profile")).json()["profile"]["basic_info"] == {
+            "name": "local"
+        }
+
+        graph = (await client.get("/api/v2/memories/graph", params={"user_id": "wife"})).json()
+        labels = {node["label"] for node in graph["nodes"] if node["node_type"] == "memory"}
+        assert labels == {"wife drinks coffee", "wife walks the coffee route"}
+
+        summary = (await client.get("/api/v2/home-summary", params={"user_id": "wife"})).json()
+        assert summary["memory_count"] == 2
+
+    run_with_sdk(tmp_path, monkeypatch, scenario)
+
+
+@pytest.mark.fresh_sdk
+def test_another_users_memory_id_answers_404_for_read_update_and_delete(
+    tmp_path, monkeypatch
+) -> None:
+    async def scenario(app, sdk, client):
+        await _write_both_users(app, sdk, client)
+        [wife_memory] = [
+            memory_id
+            for memory_id, row in sdk.memories.items()
+            if row["user_id"] == "wife" and row["content"] == "wife drinks coffee"
+        ]
+        path = f"/api/v2/memories/{wife_memory}"
+
+        assert (await client.get(path)).status_code == 404
+        assert (await client.put(path, json={"content": "stolen"})).status_code == 404
+        assert (await client.delete(path)).status_code == 204
+        assert sdk.memories[wife_memory]["content"] == "wife drinks coffee"
+
+        own = await client.get(path, params={"user_id": "wife"})
+        assert own.status_code == 200 and own.json()["user_id"] == "wife"
+        updated = await client.put(path, json={"content": "wife drinks tea", "user_id": "wife"})
+        assert updated.status_code == 200
+        assert (await client.delete(path, params={"user_id": "wife"})).status_code == 204
+        assert wife_memory not in sdk.memories
+
+    run_with_sdk(tmp_path, monkeypatch, scenario)
+
+
+@pytest.mark.fresh_sdk
+def test_receipts_belong_to_the_user_that_submitted_them(tmp_path, monkeypatch) -> None:
+    async def scenario(app, sdk, client):
+        receipts = await _write_both_users(app, sdk, client)
+        wife_add = f"/api/v2/memories/additions/{receipts['add:wife']}"
+        assert (await client.get(wife_add)).status_code == 404
+        own = await client.get(wife_add, params={"user_id": "wife"})
+        assert own.status_code == 200 and own.json()["state"] == "ready"
+        local_add = f"/api/v2/memories/additions/{receipts['add:local']}"
+        assert (await client.get(local_add, params={"user_id": "wife"})).status_code == 404
+
+        # Both users used `same-request-id`; each reads only their own receipt.
+        local_write = await client.get("/api/v2/memory-writes/same-request-id")
+        wife_write = await client.get(
+            "/api/v2/memory-writes/same-request-id", params={"user_id": "wife"}
+        )
+        assert local_write.status_code == wife_write.status_code == 200
+        assert local_write.json()["memory_ids"] != wife_write.json()["memory_ids"]
+        assert (
+            await client.get("/api/v2/memory-writes/same-request-id", params={"user_id": "kid"})
+        ).status_code == 404
+        # `local` cannot name another user's namespaced queue key directly.
+        with sqlite3.connect(app.state.settings.data_dir / "one-memory.sqlite3") as db:
+            keys = [row[0] for row in db.execute("SELECT idempotency_key FROM memory_write_jobs")]
+        wife_key = next(key for key in keys if key.startswith("user:wife:"))
+        assert (await client.get(f"/api/v2/memory-writes/{wife_key}")).status_code == 404
+
+    run_with_sdk(tmp_path, monkeypatch, scenario)
+
+
+@pytest.mark.fresh_sdk
+def test_omitting_user_id_keeps_local_and_recall_and_ask_pass_it_through(
+    tmp_path, monkeypatch
+) -> None:
+    async def scenario(app, sdk, client):
+        await _write_both_users(app, sdk, client)
+        seen: list[str] = []
+        original = sdk.list_memories
+
+        def recording(**kwargs):
+            seen.append(kwargs["user_id"])
+            return original(**kwargs)
+
+        sdk.list_memories = recording  # type: ignore[method-assign]
+        # Built at startup over the real handle; rebuilt here over the stand-in.
+        app.state.recall_target = _sdk_recall_target(app, "local")
+        await client.get("/api/v2/memory-library")
+        await client.get("/api/v2/raw")
+        await client.get("/api/v2/home-summary")
+        assert set(seen) == {"local"}
+
+        monkeypatch.setattr(
+            "one_memory.api._recall_model_ready", lambda _app: _true(), raising=True
+        )
+        recalled = await client.post("/api/v1/recall", json={"query": "drinks", "user_id": "wife"})
+        assert recalled.status_code == 200, recalled.text
+        assert {hit["content"] for hit in recalled.json()["hits"]} == {
+            "wife drinks coffee",
+            "wife walks the coffee route",
+        }
+        local = await client.post("/api/v1/recall", json={"query": "drinks"})
+        assert all("wife" not in hit["content"] for hit in local.json()["hits"])
+
+        answered = await client.post(
+            "/api/v2/memories/ask", json={"question": "what does she drink", "user_id": "wife"}
+        )
+        assert answered.status_code == 200, answered.text
+        assert {memory["content"] for memory in answered.json()["memories"]} <= {
+            "wife drinks coffee",
+            "wife walks the coffee route",
+        }
+
+    run_with_sdk(tmp_path, monkeypatch, scenario)
+
+
+async def _true() -> bool:
+    return True
+
+
+@pytest.mark.fresh_sdk
+@pytest.mark.parametrize(
+    ("method", "path", "kwargs"),
+    [
+        ("get", "/api/v2/memory-library", {"params": {"nobody": "1"}}),
+        ("get", "/api/v2/home-summary", {"params": {"userid": "wife"}}),
+        ("get", "/api/v2/user-profile", {"params": {"user": "wife"}}),
+        ("get", "/api/v2/raw", {"params": {"scope": "x"}}),
+        ("get", "/api/v2/memories/graph", {"params": {"scene": "all"}}),
+        ("get", "/api/v2/memories/some-id", {"params": {"owner": "wife"}}),
+        ("delete", "/api/v2/memories/some-id", {"params": {"owner": "wife"}}),
+        ("get", "/api/v2/memory-writes/some-id", {"params": {"owner": "wife"}}),
+        ("get", f"/api/v2/memories/additions/{uuid4()}", {"params": {"owner": "wife"}}),
+        ("post", "/api/v2/memories/ask", {"json": {"question": "q", "user": "wife"}}),
+        ("post", "/api/v1/recall", {"json": {"query": "q", "user": "wife"}}),
+        ("post", "/api/v1/recall", {"json": {"query": "q"}, "params": {"user_id": "wife"}}),
+        ("post", "/api/v2/context/search", {"json": {"query": "q", "owner": "wife"}}),
+        (
+            "post",
+            "/api/v2/memories",
+            {"json": {"request_id": str(uuid4()), "text": "t", "scope": "all-apps", "u": 1}},
+        ),
+        ("put", "/api/v2/memories/some-id", {"json": {"content": "c", "owner": "wife"}}),
+    ],
+)
+def test_unknown_query_and_body_parameters_answer_422(tmp_path, method, path, kwargs) -> None:
+    async def scenario():
+        app = application(tmp_path)
+        async with app.router.lifespan_context(app), client_for(app) as client:
+            response = await client.request(method.upper(), path, **kwargs)
+            assert response.status_code == 422, response.text
+            assert response.json()["detail"] == "request_validation_failed"
+
+    asyncio.run(scenario())
+
+
+@pytest.mark.fresh_sdk
+@pytest.mark.parametrize("bad", ["a:b", "", "x" * 65, "white space", "ü", "a/b"])
+def test_a_malformed_user_id_answers_422(tmp_path, bad) -> None:
+    async def scenario():
+        app = application(tmp_path)
+        async with app.router.lifespan_context(app), client_for(app) as client:
+            responses = [
+                await client.get("/api/v2/memory-library", params={"user_id": bad}),
+                await client.get("/api/v2/user-profile", params={"user_id": bad}),
+                await client.post(
+                    "/api/v2/memories",
+                    json={
+                        "request_id": str(uuid4()),
+                        "text": "t",
+                        "scope": "all-apps",
+                        "user_id": bad,
+                    },
+                ),
+                await client.post(
+                    "/api/v2/memory-writes",
+                    json={
+                        "request_id": "r",
+                        "session_id": "s",
+                        "messages": _messages("hello"),
+                        "user_id": bad,
+                    },
+                ),
+                await client.post("/api/v1/recall", json={"query": "q", "user_id": bad}),
+            ]
+            assert [response.status_code for response in responses] == [422] * len(responses)
+
+    asyncio.run(scenario())
+
+
+def test_user_ids_and_lanes() -> None:
+    assert valid_user_id("wife") and valid_user_id("A.b_c-9")
+    assert not valid_user_id("a:b") and not valid_user_id("wife\n")
+    assert lane_key("local", "hermes") == "hermes"
+    assert lane_key("wife", "hermes") == "hermes@wife"
+    assert split_lane("hermes@wife") == ("wife", "hermes")
+    assert split_lane("hermes") == ("local", "hermes")
+    # `local` keeps the session identity it always had; another user's never equals it.
+    assert internal_session_id_for("s-1", "hermes") == "s-1"
+    assert internal_session_id_for("s-1", "hermes", "wife") != "s-1"
+    assert internal_session_id_for("s-1", "hermes", "wife") != internal_session_id_for(
+        "s-1", "hermes", "kid"
+    )
+
+
+def _hermes_state_db(root: Path, rows: list[tuple[int, str, str, str]]) -> None:
+    root.mkdir(parents=True, exist_ok=True)
+    connection = sqlite3.connect(root / "state.db")
+    connection.execute(
+        "CREATE TABLE messages (id INTEGER PRIMARY KEY, session_id TEXT, role TEXT,"
+        " content TEXT, timestamp REAL, tool_name TEXT, active INTEGER)"
+    )
+    connection.executemany(
+        "INSERT INTO messages VALUES (?, ?, ?, ?, 1790000000, NULL, 1)",
+        rows,
+    )
+    connection.commit()
+    connection.close()
+
+
+def _hermes_homes(home: Path) -> tuple[Path, Path]:
+    hermes = home / ".hermes"
+    wife = hermes / "profiles" / "wife"
+    # The same Hermes session id in both profiles: identities must not merge.
+    _hermes_state_db(
+        hermes,
+        [(1, "session-1", "user", "I am the husband"), (2, "session-1", "assistant", "hi")],
+    )
+    _hermes_state_db(
+        wife,
+        [(1, "session-1", "user", "I am the wife"), (2, "session-1", "assistant", "hello")],
+    )
+    return hermes, wife
+
+
+def _scan_texts(root: Path, home: Path) -> list[str]:
+    adapter = create_hermes_adapter(
+        AuthorizedSourceRoot.grant(
+            source_id=HERMES_DESCRIPTOR.source_id, root=root, user_confirmed=True, home=home
+        )
+    )
+
+    async def collect() -> list[str]:
+        return [
+            event.evidence.verbatim
+            async for event in adapter.scan(ScanOptions(cursors={}))
+            if isinstance(event, ScannedEvidence)
+        ]
+
+    return asyncio.run(collect())
+
+
+def test_a_hermes_root_scan_never_reads_its_profiles(tmp_path) -> None:
+    hermes, wife = _hermes_homes(tmp_path)
+    (hermes / "profiles" / "wife" / "sessions").mkdir()
+    (hermes / "profiles" / "wife" / "sessions" / "s.jsonl").write_text(
+        json.dumps({"type": "message", "role": "user", "content": "wife's transcript"}) + "\n"
+    )
+
+    assert _scan_texts(hermes, tmp_path) == ["I am the husband", "hi"]
+    assert _scan_texts(wife, tmp_path) == ["I am the wife", "hello", "wife's transcript"]
+
+
+def _jobs(database: Path) -> list[sqlite3.Row]:
+    with sqlite3.connect(database) as db:
+        db.row_factory = sqlite3.Row
+        return db.execute(
+            "SELECT session_key,source_key,idempotency_key,state,payload_json "
+            "FROM memory_write_jobs ORDER BY queue_seq"
+        ).fetchall()
+
+
+@pytest.mark.fresh_sdk
+def test_two_hermes_grants_each_read_their_own_root_into_separate_batches(tmp_path) -> None:
+    async def scenario():
+        home = tmp_path / "home"
+        _hermes, wife = _hermes_homes(home)
+        app = application(tmp_path, home=home)
+        async with app.router.lifespan_context(app), client_for(app) as client:
+            location = (await client.post("/api/v1/sources/hermes/location")).json()
+            local = await client.post(
+                "/api/v1/sources/connect",
+                json={
+                    "source_ids": ["hermes"],
+                    "request_id": str(uuid4()),
+                    "source_roots": {"hermes": location["root_id"]},
+                    "enable_sync": True,
+                },
+            )
+            assert local.status_code == 200, local.text
+            profile = await client.post(
+                "/api/v1/sources/connect",
+                json={
+                    "source_ids": ["hermes"],
+                    "request_id": str(uuid4()),
+                    "user_id": "wife",
+                    "root": str(wife),
+                },
+            )
+            assert profile.status_code == 200, profile.text
+            [wife_job] = profile.json()["jobs"]
+            assert wife_job["user_id"] == "wife"
+            jobs = [*local.json()["jobs"], wife_job]
+            for job in jobs:
+                finished = await app.state.import_supervisor.wait(job["job_id"])
+                assert finished.status == "completed", finished
+
+            rows = [row for row in _jobs(app.state.settings.data_dir / "one-memory.sqlite3")]
+            assert rows, "history import wrote no batches"
+            by_user: dict[str, list[str]] = {}
+            for row in rows:
+                records = json.loads(row["payload_json"])["input_records"]
+                users = {record["user_id"] for record in records}
+                assert len(users) == 1, "a batch mixed users"
+                [user] = users
+                by_user.setdefault(user, []).extend(record["verbatim"] for record in records)
+                assert row["source_key"] == lane_key(user, "hermes")
+                assert row["session_key"] == internal_session_id_for("session-1", "hermes", user)
+            assert sorted(by_user["local"]) == ["I am the husband", "hi"]
+            assert sorted(by_user["wife"]) == ["I am the wife", "hello"]
+
+            grants = (await client.get("/api/v1/source-grants")).json()["grants"]
+            assert {(grant["user_id"], grant["source_id"]) for grant in grants} >= {
+                ("local", "hermes"),
+                ("wife", "hermes"),
+            }
+            # The App's view stays `local`'s.
+            listed = (await client.get("/api/v1/import-jobs")).json()
+            assert {job["user_id"] for job in listed} == {"local"}
+            wife_jobs = (await client.get("/api/v1/import-jobs", params={"user_id": "wife"})).json()
+            assert {job["user_id"] for job in wife_jobs} == {"wife"}
+            assert all(
+                item["source_id"] != "hermes@wife"
+                for item in (await client.get("/api/v1/live-sync")).json()
+            )
+
+            revoked = await client.post("/api/v1/sources/hermes/revoke", params={"user_id": "wife"})
+            assert revoked.status_code == 200 and revoked.json()["revoked_at"] is not None
+            local_grant = app.state.source_grant_store.receipt(user_id="local", source_id="hermes")
+            assert local_grant.revoked_at is None
+
+    asyncio.run(scenario())
+
+
+@pytest.mark.fresh_sdk
+def test_only_hermes_takes_another_user_or_a_custom_root(tmp_path) -> None:
+    async def scenario():
+        home = tmp_path / "home"
+        (home / ".claude" / "projects").mkdir(parents=True)
+        app = application(tmp_path, home=home)
+        async with app.router.lifespan_context(app), client_for(app) as client:
+            for extra in ({"user_id": "wife"}, {"root": str(home / ".claude" / "projects")}):
+                response = await client.post(
+                    "/api/v1/sources/connect",
+                    json={
+                        "source_ids": ["claude-code"],
+                        "request_id": str(uuid4()),
+                        "source_roots": {},
+                        **extra,
+                    },
+                )
+                assert response.status_code == 422, response.text
+                assert response.json()["detail"] == "user_id_requires_multi_user_source"
+            revoke = await client.post(
+                "/api/v1/sources/claude-code/revoke", params={"user_id": "wife"}
+            )
+            assert revoke.status_code == 422
+            # Hermes too: another user must name their own root.
+            rootless = await client.post(
+                "/api/v1/sources/connect",
+                json={
+                    "source_ids": ["hermes"],
+                    "request_id": str(uuid4()),
+                    "source_roots": {},
+                    "user_id": "wife",
+                },
+            )
+            assert rootless.status_code == 422
+            assert rootless.json()["detail"] == "source_root_required_for_user"
+
+    asyncio.run(scenario())
+
+
+def _record(user_id: str, message_id: str, role: str, text: str):
+    return materialize_write_input(
+        EvidenceInput(
+            user_id=user_id,
+            source_id="hermes",
+            conversation_id="session-1",
+            message_id=message_id,
+            role=role,
+            written_by=WrittenBy.PERSON if role == "user" else WrittenBy.AGENT,
+            created_at=datetime(2026, 10, 1, tzinfo=UTC),
+            verbatim=text,
+            origin_meta={EvidenceOrigin.FILE: OriginMetadata()},
+        )
+    )
+
+
+def test_assembling_rows_and_batches_are_keyed_by_user(tmp_path) -> None:
+    """Two users' `state.db` share a relative file key and a session id.
+
+    Finishing one user's file must neither take nor drop the other user's
+    staged turns, and the batches it writes hold one user each.
+    """
+    path = tmp_path / "one-memory.sqlite3"
+    queue = DurableWriteQueue(path, sdk_authority=True)
+    file_key = "f" * 64
+    from one_memory.database_access import main_database
+
+    with main_database(path).write(row_factory=sqlite3.Row) as connection, connection:
+        for user in ("local", "wife"):
+            for index, role in enumerate(("user", "assistant")):
+                queue.stage_sdk_input(
+                    _record(user, f"m{index}", role, f"{user} {role}"),
+                    file_key=file_key,
+                    priority="normal",
+                    connection=connection,
+                )
+        queue.finalize_sdk_file(
+            user_id="local", source_id="hermes", file_key=file_key, connection=connection
+        )
+
+    rows = _jobs(path)
+    pending = [row for row in rows if row["state"] == "pending"]
+    assembling = [row for row in rows if row["state"] == "assembling"]
+    assert [
+        {record["user_id"] for record in json.loads(row["payload_json"])["input_records"]}
+        for row in pending
+    ] == [{"local"}]
+    assert [split_lane(row["source_key"]) for row in assembling] == [("wife", "hermes")]
+    assert pending[0]["session_key"] != assembling[0]["session_key"]
+
+
+def _observed(response) -> tuple[int, bytes]:
+    return response.status_code, response.content
+
+
+@pytest.mark.fresh_sdk
+def test_another_users_id_answers_exactly_like_an_id_that_does_not_exist(
+    tmp_path, monkeypatch
+) -> None:
+    """Comparing responses must not tell whether another user's id exists."""
+
+    async def scenario(app, sdk, client):
+        receipts = await _write_both_users(app, sdk, client)
+        [wife_memory] = [
+            memory_id
+            for memory_id, row in sdk.memories.items()
+            if row["user_id"] == "wife" and row["content"] == "wife drinks coffee"
+        ]
+        before = {key: dict(row) for key, row in sdk.memories.items()}
+        with sqlite3.connect(app.state.settings.data_dir / "one-memory.sqlite3") as db:
+            keys = [row[0] for row in db.execute("SELECT idempotency_key FROM memory_write_jobs")]
+        wife_key = next(key for key in keys if key.startswith("user:wife:"))
+
+        pairs = [
+            ("GET", f"/api/v2/memories/{wife_memory}", "/api/v2/memories/no-such-memory", {}),
+            (
+                "PUT",
+                f"/api/v2/memories/{wife_memory}",
+                "/api/v2/memories/no-such-memory",
+                {"json": {"content": "stolen"}},
+            ),
+            ("DELETE", f"/api/v2/memories/{wife_memory}", "/api/v2/memories/no-such-memory", {}),
+            (
+                "GET",
+                f"/api/v2/memories/additions/{receipts['add:wife']}",
+                f"/api/v2/memories/additions/{uuid4()}",
+                {},
+            ),
+            ("GET", f"/api/v2/memory-writes/{wife_key}", "/api/v2/memory-writes/no-such-id", {}),
+        ]
+        for method, theirs, missing, kwargs in pairs:
+            other = await client.request(method, theirs, **kwargs)
+            absent = await client.request(method, missing, **kwargs)
+            assert _observed(other) == _observed(absent), (method, theirs)
+            assert other.status_code == (204 if method == "DELETE" else 404)
+        assert sdk.memories == before
+
+    run_with_sdk(tmp_path, monkeypatch, scenario)

memhub/apps/core/tests/test_policy_api_wiring.py

@@ -173,12 +173,14 @@ def test_first_import_gate_waits_for_every_selected_history_job() -> None:
     jobs = [
         SimpleNamespace(
             operation="history",
+            user_id="local",
             source_id="cursor",
             created_at=datetime(2026, 10, 18, tzinfo=UTC),
             status="completed",
         ),
         SimpleNamespace(
             operation="history",
+            user_id="local",
             source_id="codex",
             created_at=datetime(2026, 10, 18, tzinfo=UTC),
             status="running",

memhub/apps/core/tests/test_recall_raw_fallback.py

@@ -210,7 +210,7 @@ def test_recall_route_falls_back_and_keeps_the_raw_grant(tmp_path, monkeypatch):
     async def no_model(_app):
         return False
 
-    async def raw(_app, query, *, limit, sources):
+    async def raw(_app, query, *, limit, sources, user_id="local"):
         searched.append(sources)
         return (_raw("r1", "上周决定用 Postgres", source="cursor"),)
 

memhub/apps/core/tests/test_sdk_data_plane_migration.py

@@ -1353,7 +1353,7 @@ def test_the_sdk_memory_map_draws_sdk_memories_without_legacy_tables(tmp_path) -
             original = app.state.sdk_client_handle
             app.state.sdk_client_handle = Handle()
 
-            async def graph_snapshot(*, limit):
+            async def graph_snapshot(*, limit, user_id="local"):
                 del limit
                 return HyGraphSnapshot(
                     nodes=(HyGraphNode(node_id="t1", content="topic", layer="TOPIC"),),

memhub/apps/core/tests/test_sdk_migration_bundle.py

@@ -91,7 +91,7 @@ def test_bundle_contains_management_sdk_config_and_verified_manifest(tmp_path):
     assert manifest["bundle_id"] == "migration-1"
     assert manifest["sdk_namespace"] == "governed-v2"
     assert manifest["runtime_build"] == "build-13"
-    assert manifest["snapshot_sdk_version"] == "2.0.8"
+    assert manifest["snapshot_sdk_version"] == "2.0.9"
     paths = {item["path"]: item for item in manifest["files"]}
     assert {
         "management/one-memory.sqlite3",

memhub/apps/core/tests/test_v2_authority_and_call_events.py

@@ -57,10 +57,10 @@ def client_for(app, headers=None):
 
 
 class Answerer:
-    async def answer(self, question):
+    async def answer(self, question, *, user_id="local"):
         return SdkMemoryAnswer(answer="真实数据。", memories=(MEMORY,), mode="extractive")
 
-    async def answer_stream(self, question):
+    async def answer_stream(self, question, *, user_id="local"):
         yield {"type": "recall", "memories": (MEMORY,)}
         yield {
             "type": "final",

memhub/apps/core/uv.lock

@@ -991,7 +991,7 @@ wheels = [
 
 [[package]]
 name = "hy-memory"
-version = "2.0.8"
+version = "2.0.9"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "chromadb" },
@@ -1008,9 +1008,9 @@ dependencies = [
     { name = "scikit-learn" },
     { name = "tiktoken" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/21/15/83ca83b649ead238f186fdb33efdc73615fbba279efbbdca3857c1a86beb/hy_memory-2.0.8.tar.gz", hash = "sha256:0fa0ed1a1160b389a3d7c13a9fb6528b140d7ec23cf49e65f1be5974820a576b", size = 856025, upload-time = "2026-10-28T07:35:47.783Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/48/28/e01fd9a92cd090c8236df0dd06dae049c5de66345e5bd9653497ed56125e/hy_memory-2.0.9.tar.gz", hash = "sha256:5f55ea92044ced19a00b4e56c565215adedd3439a1bdb9cfe9ffe42387c8f1a1", size = 858323, upload-time = "2026-11-01T00:29:49.914Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/11/6a/c699cac0ff06bd089fceeff90446130763d8187e649dbdb0313014491a10/hy_memory-2.0.8-py3-none-any.whl", hash = "sha256:68bd789df24e51e64bb231e97a8d24ee89b4aa5f9ba14c3e0824589e9213bf7c", size = 921119, upload-time = "2026-10-28T07:35:45.571Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/44/4368555a408a58daf902c81b62184a3c99f12abd45d4f1144bf79f171a6f/hy_memory-2.0.9-py3-none-any.whl", hash = "sha256:95dbea232d2041f9b0ea3d08dd751603c2a5c0e4422b97e74f664e6ca913549e", size = 923250, upload-time = "2026-11-01T00:29:47.814Z" },
 ]
 
 [[package]]
@@ -1837,7 +1837,7 @@ requires-dist = [
     { name = "fastapi", specifier = ">=0.141.1" },
     { name = "filelock", specifier = ">=3.32.4" },
     { name = "httpx", specifier = ">=0.28.1" },
-    { name = "hy-memory", specifier = "==2.0.8" },
+    { name = "hy-memory", specifier = "==2.0.9" },
     { name = "keyring", specifier = ">=25.7.0" },
     { name = "numpy", specifier = "==2.5.2" },
     { name = "onnxruntime", specifier = "==1.23.2" },

memhub/apps/web/src/memory-processing-labels.ts

@@ -184,6 +184,9 @@ const SCAN_ISSUE_LABELS: Record<string, string> = {
   tool_content_excluded: "工具输出不计入原文",
   internal_reasoning_excluded: "模型的内部思考不计入原文",
   not_written_by_person: "不是本人写的，未计入",
+  injected_machine_notice: "来源程序自己发的通知，未计入",
+  voice_placeholder: "语音没有转成文字，未计入",
+  probe: "连通性检查消息，未计入",
   source_busy: "来源正被占用，稍后会自动补读",
   unsupported_schema: "这个来源的文件格式暂不支持",
   budget_exceeded: "超出本次处理上限，其余未读",

memhub/docs/ARCHITECTURE.md

@@ -282,7 +282,7 @@ Support is declared per immutable GitHub Release manifest:
 - Linux x64/arm64 Runtime on native GHA runners.
 - macOS arm64/x64 App/Runtime: Developer ID, notarization, native acceptance.
 - Windows is not supported.
-- `hy-memory==2.0.8` is the current stable SDK dependency; the release
+- `hy-memory==2.0.9` is the current stable SDK dependency; the release
   lock and [`../MEMORY.md`](../MEMORY.md) are authoritative when the pin changes.
 
 See [`INSTALLATION.md`](INSTALLATION.md).

memhub/docs/cli/EXPERIENCE.md

@@ -53,6 +53,26 @@ node apps/cli/dist/one-memory.cjs sources connect --all-detected --import
 （如 `DSH_HOME`）的来源必须加 `--confirm-overrides`，否则被拒绝（或在
 `--all-detected` 时跳过并列出）。用同一个 `--request-id <uuid>` 重试不会重复导入。
 
+### 接入 Hermes 的另一个 profile（多用户）
+
+一个 Runtime 可以给多个用户记忆。**用户由读的是哪个目录决定**，不看会话或平台账号：
+`~/.hermes/state.db` 属于默认用户 `local`，`~/.hermes/profiles/<name>/state.db` 是另一个用户。
+
+```bash
+# 把 wife 这个 Hermes profile 接成用户 wife：授权 + 导入历史，一步完成
+node apps/cli/dist/one-memory.cjs sources connect hermes --user-id wife --root ~/.hermes/profiles/wife
+# 也可以显式加 --import，效果相同；可带 --request-id <uuid> 幂等重试
+node apps/cli/dist/one-memory.cjs sources connect hermes --user-id wife --root ~/.hermes/profiles/wife --import --json
+
+# 列出来源；其他用户的连接单独列在「其他用户的连接」下，--json 里在 grants 字段
+node apps/cli/dist/one-memory.cjs sources list
+```
+
+- 只有 `hermes` 能带 `--user-id`（非 `local`）或 `--root`；其他来源带了会退 2，什么都不发。
+- `--root` 里的 `~` 由 CLI 展开为主目录，相对路径按当前目录解析成绝对路径。
+- 扫描 `~/.hermes` 时会排除 `profiles/**`，所以 `local` 不会读到 wife 的 `state.db`。
+- App 不变：它只读写 `local`，其他来源也都固定是 `local`。
+
 ## 3. 看导入进度
 
 ```bash
@@ -105,6 +125,23 @@ node apps/cli/dist/one-memory.cjs write status <request_id> --json
 
 两条都返回 request id；它是幂等键，结果未知时用**同一个** `--request-id` 重试不会重复写入。
 
+### 按用户读写（`--user-id`）
+
+记忆命令都接受全局 `--user-id <id>`（默认 `local`，即 App 看到的那份）：
+`memories list/show/add/add-status/update/delete`、`recall`、`profile`、`write submit/status`。
+
+```bash
+node apps/cli/dist/one-memory.cjs memories list --user-id wife --json
+node apps/cli/dist/one-memory.cjs memories add "我更喜欢喝绿茶" --user-id wife
+node apps/cli/dist/one-memory.cjs recall "喝什么茶" --user-id wife
+node apps/cli/dist/one-memory.cjs profile --user-id wife
+node apps/cli/dist/one-memory.cjs write status <request_id> --user-id wife
+```
+
+- 用户 ID 只能是 1–64 位字母、数字、`.`、`_`、`-`；其他写法退 2，不连 Runtime。
+- 不带 `--user-id`（或写 `--user-id local`）时请求里不出现 `user_id`，和以前完全一样。
+- 用户之间互相隔离：读、改、删另一个用户的记忆 ID 返回「未找到」；write 的 request id 也只属于提交它的用户。
+
 ## 8. 召回（recall）
 
 `recall` 两条路径（raw / memory）此前已真机验证。文本输出会标明每条是「原始对话摘录」

memhub/packages/runtime-control/src/diagnostic-log.test.ts

@@ -90,6 +90,7 @@ describe("diagnostic log", () => {
     expect(await readDiagnosticEvents(dataDir)).toEqual({
       events: [],
       invalid_lines: 0,
+      unrecognized_lines: 0,
     });
   });
 
@@ -496,6 +497,52 @@ describe("diagnostic log", () => {
     expect(await readdir(outputDir)).toEqual([]);
   });
 
+  it("counts only non-JSON lines as damaged and leaves older records out", async () => {
+    const dataDir = await root();
+    const logger = new DiagnosticLogger({ dataDir, service: "core" });
+    await logger.log({
+      level: "info",
+      component: "core.service",
+      event: "core.ready",
+      outcome: "succeeded",
+    });
+    const older = {
+      schema_version: 0,
+      service: "desktop",
+      component: "desktop.lifecycle",
+      event: "desktop.started",
+    };
+    await writeFile(
+      path.join(dataDir, "logs", "diagnostic-desktop.jsonl"),
+      [
+        JSON.stringify(older),
+        JSON.stringify({ ...older, schema_version: 1 }),
+        JSON.stringify(["not", "an", "event"]),
+        "",
+      ].join("\n"),
+      { mode: 0o600 },
+    );
+
+    const result = await readDiagnosticEvents(dataDir);
+    expect(result.events.map((event) => event.event)).toEqual(["core.ready"]);
+    expect(result.invalid_lines).toBe(0);
+    expect(result.unrecognized_lines).toBe(3);
+    // An older build's records do not stop a support export either.
+    const target = path.join(await root(), "bundle.json");
+    await expect(
+      exportDiagnosticBundle({ dataDir, target, productVersion: "1.2.3" }),
+    ).resolves.toEqual({ event_count: 1, invalid_lines: 0 });
+
+    await writeFile(
+      path.join(dataDir, "logs", "diagnostic-cli.jsonl"),
+      '{"schema_version": 1, "event": \n',
+      { mode: 0o600 },
+    );
+    const damaged = await readDiagnosticEvents(dataDir);
+    expect(damaged.invalid_lines).toBe(1);
+    expect(damaged.unrecognized_lines).toBe(3);
+  });
+
   it("fails closed when the diagnostic directory cannot be read", async () => {
     if (process.platform === "win32") return;
     const dataDir = await root();

memhub/packages/runtime-control/src/diagnostic-log.ts

@@ -429,10 +429,18 @@ export async function readDiagnosticEvents(
     errorsOnly?: boolean;
     tail?: number;
   } = {},
-): Promise<{ events: DiagnosticEventV1[]; invalid_lines: number }> {
+): Promise<{
+  events: DiagnosticEventV1[];
+  invalid_lines: number;
+  unrecognized_lines: number;
+}> {
   const files = await diagnosticLogFiles(dataDir);
   let events: DiagnosticEventV1[] = [];
+  // Only a line that is not JSON at all is damage. A whole JSON line this
+  // reader does not accept (an older build's record, another schema) is left
+  // out and counted apart, so an upgrade does not read as a broken log.
   let invalidLines = 0;
+  let unrecognizedLines = 0;
   for (const file of files) {
     const content = await readRegularFileNoFollow(
       file,
@@ -440,11 +448,17 @@ export async function readDiagnosticEvents(
     );
     for (const line of content.split("\n")) {
       if (!line) continue;
+      let value: unknown;
       try {
-        const item = parseDiagnosticEvent(JSON.parse(line));
-        events.push(item);
+        value = JSON.parse(line);
       } catch {
         invalidLines += 1;
+        continue;
+      }
+      try {
+        events.push(parseDiagnosticEvent(value));
+      } catch {
+        unrecognizedLines += 1;
       }
     }
   }
@@ -480,7 +494,11 @@ export async function readDiagnosticEvents(
     return !(filter.errorsOnly && item.level !== "error");
   });
   const tail = Math.min(Math.max(filter.tail ?? 200, 1), 5_000);
-  return { events: events.slice(-tail), invalid_lines: invalidLines };
+  return {
+    events: events.slice(-tail),
+    invalid_lines: invalidLines,
+    unrecognized_lines: unrecognizedLines,
+  };
 }
 
 export async function exportDiagnosticBundle(options: {

memhub/scripts/generate_sbom.py

@@ -36,7 +36,7 @@ MEMMY_LICENSE = (
 MEMMY_PROVENANCE = MEMMY_LICENSE.with_name("PROVENANCE.md")
 MEMMY_PIN = "25acd5f29fc7df73b943713abe61f1b2f9839887"
 HY_MEMORY_NAME = "hy-memory"
-HY_MEMORY_VERSION = "2.0.8"
+HY_MEMORY_VERSION = "2.0.9"
 
 
 class SbomError(RuntimeError):

memhub/scripts/install-docs.test.mjs

@@ -12,7 +12,7 @@ test("installation docs disclose publication and lifecycle boundaries", async ()
     "utf8",
   );
 
-  assert.match(readme, /`hy-memory==2\.0\.8`/);
+  assert.match(readme, /`hy-memory==2\.0\.9`/);
   assert.match(readme, /`v1\.0\.5` 的发行目标是 Linux x64\/arm64 Runtime/);
   assert.match(readme, /Linux 制品由 GitHub Actions/);
   assert.match(readme, /macOS.*Developer ID 签名、公证/s);

memhub/scripts/test_generate_sbom.py

@@ -13,7 +13,7 @@ spec.loader.exec_module(module)
 
 
 def test_hy_memory_notice_comes_from_the_installed_dist_info(tmp_path: Path) -> None:
-    dist_info = tmp_path / "hy_memory-2.0.8.dist-info"
+    dist_info = tmp_path / "hy_memory-2.0.9.dist-info"
     licenses = dist_info / "licenses"
     licenses.mkdir(parents=True)
     (dist_info / "METADATA").write_text(
@@ -21,7 +21,7 @@ def test_hy_memory_notice_comes_from_the_installed_dist_info(tmp_path: Path) ->
             (
                 "Metadata-Version: 2.4",
                 "Name: hy-memory",
-                "Version: 2.0.8",
+                "Version: 2.0.9",
                 "License-Expression: Apache-2.0",
                 "",
             )
@@ -38,7 +38,7 @@ def test_hy_memory_notice_comes_from_the_installed_dist_info(tmp_path: Path) ->
     notice = module.hy_memory_notice(dist_info)
 
     assert notice["name"] == "hy-memory"
-    assert notice["version"] == "2.0.8"
+    assert notice["version"] == "2.0.9"
     assert notice["license"] == "Apache-2.0"
     assert notice["copyright"] == "Copyright (C) 2026 Tencent. All rights reserved."
     assert "Licensed under the Apache License" in notice["text"]

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "hy-memory"
-version = "2.0.8"
+version = "2.0.9"
 description = "HY Memory - Industrial-grade dual-system cognitive memory framework"
 readme = "README.md"
 requires-python = ">=3.9"

tests/test_basic_info_placeholder_guard.py

@@ -0,0 +1,175 @@
+"""P0-1：占位符不得污染 L0 基础画像。
+
+弱模型（qwen3.5-2b）在用户什么都没说时把 basic_info 填成 {"name": "用户"} /
+"The user" / "N/A"，每个占位符都 supersede 掉真实值，演化链变成
+费天翔 → Fei → 用户 → The user。防御：(a) upsert 占位符守卫 + 读侧愈合；
+(c) prompt 硬约束。验收 2（assistant 引用 "name is 用户"）由占位符守卫单独兜住。
+本文件覆盖改进建议第 10 节验收 1–4。
+"""
+import asyncio
+from datetime import datetime, timedelta
+
+import pytest
+
+from hy_memory.agent.extractor import EXTRACT_FEW_SHOT_EN
+from hy_memory.agent.prompts_zh import EXTRACT_FEW_SHOT_ZH
+from hy_memory.agent.tools.basic_profile import (
+    _assemble_full_kv_from_nodes,
+    is_placeholder_value,
+    upsert_basic_profile,
+)
+from hy_memory.models.memory import MemoryLayer, MemoryNode, MemoryStatus
+
+FIELDS = ["name", "age", "location", "occupation", "employer"]
+
+
+class FakeStore:
+    def __init__(self, nodes=None):
+        self.nodes = list(nodes or [])
+
+    async def list_by_user(self, **kwargs):
+        return [n for n in self.nodes if n.user_id == kwargs.get("user_id")]
+
+    async def get_by_id(self, node_id):
+        return next((n for n in self.nodes if n.node_id == node_id), None)
+
+    async def upsert(self, node):
+        self.nodes.append(node)
+
+    async def update_payload(self, node_id, updates):
+        node = await self.get_by_id(node_id)
+        for key, value in updates.items():
+            if key == "status" and isinstance(value, str):
+                value = MemoryStatus(value)
+            setattr(node, key, value)
+
+
+class FakeEmbed:
+    async def embed_queued(self, text):
+        return [0.1] * 8
+
+
+def l0_versions(store):
+    return [n for n in store.nodes if n.layer == MemoryLayer.L0_BASIC_INFO]
+
+
+def write_basic_info(store, llm_basic_info):
+    """模拟 writer 的 basic_info 段：LLM 给出的非空 basic_info 直接 upsert。"""
+    if not llm_basic_info:
+        return None
+    return asyncio.run(upsert_basic_profile(
+        user_id="u1", kv=llm_basic_info, vector_store=store,
+        embed_service=FakeEmbed(), allowed_fields=FIELDS,
+    ))
+
+
+# ── 验收 1：寒暄不产生新的 L0 版本 ──
+
+def test_small_talk_creates_no_l0_version():
+    # [user]: 你好 / [assistant]: 我可以帮你。弱模型输出 {"name": "用户"}
+    store = FakeStore()
+    write_basic_info(store, {"name": "用户"})
+    write_basic_info(store, {})
+    assert l0_versions(store) == []
+
+
+def test_placeholder_only_upsert_writes_nothing():
+    store = FakeStore()
+    for kv in ({"name": "用户"}, {"name": "The user"}, {"name": "N/A"},
+               {"location": "unknown", "occupation": "未提供"}, {"name": "name"}):
+        res = write_basic_info(store, kv)
+        assert res.success and not res.changed
+    assert l0_versions(store) == []
+
+
+# ── 验收 2：assistant 引用 "name is 用户" 不写入（仅靠占位符守卫）──
+
+def test_assistant_quoting_placeholder_does_not_write():
+    # [assistant]: 那条 The user's name is 用户. 更说明问题
+    store = FakeStore()
+    for kv in ({"name": "用户"}, {"name": "The user"}, {"name": "The user."}):
+        res = write_basic_info(store, kv)
+        assert res.success and not res.changed
+    assert l0_versions(store) == []
+
+
+# ── 验收 3：用户自报姓名 ──
+
+def test_user_states_name():
+    # [user]: 我叫费天翔，常住杭州，做 AI 基础设施。
+    store = FakeStore()
+    res = write_basic_info(store, {
+        "name": "费天翔", "location": "杭州", "occupation": "AI 基础设施从业者",
+    })
+    assert res.success and res.created
+    assert l0_versions(store)[0].custom["basic_info_kv"] == {
+        "name": "费天翔", "location": "杭州", "occupation": "AI 基础设施从业者",
+    }
+
+
+# ── 验收 4：已有真实值不被后续占位符覆盖 ──
+
+def test_existing_name_not_overwritten_by_placeholder():
+    store = FakeStore()
+    write_basic_info(store, {"name": "费天翔"})
+    assert len(l0_versions(store)) == 1
+    for kv in ({"name": "用户"}, {"name": "The user."}, {"name": "N/A"}):
+        res = write_basic_info(store, kv)
+        assert not res.changed
+    assert len(l0_versions(store)) == 1
+    assert _assemble_full_kv_from_nodes(list(store.nodes)) == {"name": "费天翔"}
+
+
+def test_mixed_kv_keeps_real_fields_and_drops_placeholders():
+    store = FakeStore()
+    res = write_basic_info(store, {"name": "用户", "location": "杭州"})
+    assert res.diff_kv == {"location": "杭州"}
+
+
+def test_already_poisoned_chain_heals_on_read():
+    now = datetime.utcnow()
+
+    def node(nid, kv, days_ago):
+        return MemoryNode(
+            node_id=nid, user_id="u1", agent_id="",
+            layer=MemoryLayer.L0_BASIC_INFO, content="profile",
+            custom={"basic_info_kv": kv},
+            gmt_created=now - timedelta(days=days_ago),
+        )
+
+    chain = [
+        node("a", {"name": "费天翔"}, 4), node("b", {"name": "Fei"}, 3),
+        node("c", {"name": "用户"}, 2), node("d", {"name": "The user"}, 1),
+    ]
+    assert _assemble_full_kv_from_nodes(chain) == {"name": "Fei"}
+
+
+@pytest.mark.parametrize("value", [
+    "", "  ", "用户", "用户。", "The user", "the user.", "THE USER", "user",
+    "The user's", "n/a", "N/A", "na", "None", "null", "unknown", "未提供",
+    "未知", "不详", "-", "?", "name", " Name ", "姓名",
+])
+def test_placeholder_values(value):
+    assert is_placeholder_value("name", value)
+
+
+@pytest.mark.parametrize("field,value", [
+    ("name", "费天翔"), ("name", "Fei"), ("name", "Username42"),
+    ("age", 30), ("age", 0), ("location", "杭州"), ("employer", "Tencent"),
+])
+def test_real_values_are_not_placeholders(field, value):
+    assert not is_placeholder_value(field, value)
+
+
+# ── (c) prompt 硬约束 ──
+
+def test_prompts_carry_hard_rule_and_small_talk_counter_example():
+    from hy_memory.agent import extractor, prompts_zh
+
+    en_rules = [v for v in vars(extractor).values()
+                if isinstance(v, str) and "HARD RULE: fill a field ONLY" in v]
+    zh_rules = [v for v in vars(prompts_zh).values()
+                if isinstance(v, str) and "**硬性规则**" in v]
+    assert len(en_rules) >= 2 and len(zh_rules) >= 2
+    assert "small talk has no basic_info" in EXTRACT_FEW_SHOT_EN
+    assert "寒暄没有 basic_info" in EXTRACT_FEW_SHOT_ZH

tests/test_reembed.py

@@ -0,0 +1,642 @@
+"""Re-embed into a new embedding space: vector stores, scan, diff and merge.
+
+The switch model: writes are paused, every vector is computed again with the
+new model into a new space, the source is never touched. The graph half lives
+in ``test_reembed_graph.py``.
+"""
+
+import hashlib
+import json
+import math
+import random
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from hy_memory import (
+    EmbeddingDimensionMismatchError,
+    MemoryConfig,
+    ReembedCancelled,
+    VectorSpace,
+    diff_spaces,
+    drop_vector_space,
+    plan_reembed,
+    reembed_text,
+    reembed_vector_space,
+    scan_vector_spaces,
+    space_identity,
+    vector_space_names,
+    verify_reembed,
+)
+from hy_memory.config import EmbedderConfig
+from hy_memory.core.embed_service import EmbedService
+from hy_memory.models.memory import MemoryLayer, MemoryNode, MemoryStatus
+from hy_memory.reembed import VECTOR_ROLES, parse_collection_name
+
+
+class FakeEmbedder:
+    """Deterministic vectors: same (model, text) -> same unit vector."""
+
+    def __init__(self, model: str, dims: int, *, wrong_width_after: int = -1):
+        self.model = model
+        self.dims = dims
+        self.calls = []
+        self.texts = 0
+        self.wrong_width_after = wrong_width_after
+
+    def vector(self, text: str):
+        seed = int(hashlib.sha256(f"{self.model}\0{text}".encode()).hexdigest()[:16], 16)
+        rng = random.Random(seed)
+        values = [rng.gauss(0.0, 1.0) for _ in range(self.dims)]
+        norm = math.sqrt(sum(v * v for v in values)) or 1.0
+        return [v / norm for v in values]
+
+    async def embed_batch(self, texts, use_cache=True):
+        self.calls.append(list(texts))
+        out = []
+        for text in texts:
+            self.texts += 1
+            vector = self.vector(text)
+            if 0 <= self.wrong_width_after < self.texts:
+                vector = vector + [0.0]
+            out.append(vector)
+        return out
+
+    async def embed(self, text, use_cache=True):
+        return self.vector(text)
+
+    async def embed_queued(self, text, use_cache=True):
+        return self.vector(text)
+
+    def get_embedding_dims(self):
+        return self.dims
+
+
+def _client(path):
+    import chromadb
+    return chromadb.PersistentClient(path=str(path))
+
+
+def _space(path, base, dims, model=None):
+    return VectorSpace(base_name=base, dims=dims, persist_directory=str(path), model=model)
+
+
+def _custom_state():
+    return {
+        "memory_kind": "canonical_state",
+        "state_key": "home_city",
+        "state_value": "Berlin",
+        "state_retrieval_text": "user lives in Berlin (home city)",
+    }
+
+
+async def _seed(path, base, dims, *, model="old-model", users=("alice", "bob"), per_user=7):
+    """Write a realistic source space through the SDK's own stores."""
+    from hy_memory.data.vector_store_chroma import ChromaVectorStore
+
+    old = FakeEmbedder(model, dims)
+    cfg = MemoryConfig()
+    cfg.vector_store.provider = "chroma"
+    cfg.vector_store.persist_directory = str(path)
+    cfg.vector_store.collection_name = base
+    cfg.vector_store.embedding_dims = dims
+    store = ChromaVectorStore(cfg)
+    await store.initialize()
+    nodes = []
+    for user in users:
+        for i in range(per_user):
+            layer = [MemoryLayer.L1_RAW, MemoryLayer.L2_FACT, MemoryLayer.L3_SUMMARY][i % 3]
+            node = MemoryNode(
+                user_id=user, agent_id="a", session_id=f"s{i % 2}",
+                layer=layer, content=f"{user} fact {i}",
+                status=MemoryStatus.SUPERSEDED if i == 4 else MemoryStatus.ACTIVE,
+                tags=["t1", f"t{i}"],
+                custom={"k": i, "nested": {"x": [1, 2]}},
+            )
+            if i == 4:
+                node.is_latest = False
+                node.superseded_by = ["later"]
+            if i == 5:
+                node.custom = _custom_state()
+            node.embedding = old.vector(node.content)
+            nodes.append(node)
+    await store.upsert_batch(nodes)
+    entities = await store._get_entity_collection()
+    entities.add(
+        ids=["e1", "e2"],
+        embeddings=[old.vector("Berlin"), old.vector("Dylan")],
+        documents=["Berlin", "Dylan"],
+        metadatas=[{"user_id": "alice", "label": "GPE"}, {"user_id": "bob", "label": "PERSON"}],
+    )
+    await store.close()
+    client = _client(path)
+    names = vector_space_names(_space(path, base, dims))
+    for role, doc in (("coding_keys", "fix the flaky test"), ("working_facts_cv", "repo: uses uv")):
+        coll = client.get_or_create_collection(names[role], metadata={"hnsw:space": "cosine"})
+        coll.add(
+            ids=[f"{role}:0"],
+            embeddings=[old.vector(doc)],
+            documents=[doc],
+            metadatas=[{"user_id": "alice", "layer": "l1_raw",
+                        "custom": json.dumps({"coding_key_text": doc})}],
+        )
+    return names
+
+
+def _snapshot(path, names):
+    client = _client(path)
+    out = {}
+    for role, name in names.items():
+        try:
+            coll = client.get_collection(name)
+        except Exception:
+            continue
+        got = coll.get(include=["documents", "metadatas", "embeddings"])
+        out[role] = sorted(
+            (rid, doc, json.dumps(meta, sort_keys=True), [round(float(x), 6) for x in emb])
+            for rid, doc, meta, emb in zip(got["ids"], got["documents"], got["metadatas"], got["embeddings"])
+        )
+    return hashlib.sha256(json.dumps(out, sort_keys=True).encode()).hexdigest()
+
+
+def _records(path, name):
+    got = _client(path).get_collection(name).get(include=["documents", "metadatas", "embeddings"])
+    return {
+        rid: (doc, meta, list(emb))
+        for rid, doc, meta, emb in zip(got["ids"], got["documents"], got["metadatas"], got["embeddings"])
+    }
+
+
+# ----------------------------------------------------------------------
+# naming
+# ----------------------------------------------------------------------
+
+class TestVectorSpaceNames:
+    @pytest.mark.parametrize("dims", [512, None])
+    async def test_names_equal_what_the_stores_open(self, tmp_path, dims):
+        from hy_memory.coding.store import CodingMemoryStore
+        from hy_memory.data.vector_store_chroma import ChromaVectorStore
+
+        cfg = MemoryConfig()
+        cfg.vector_store.provider = "chroma"
+        cfg.vector_store.persist_directory = str(tmp_path / "v")
+        cfg.vector_store.collection_name = "memhub_abc123"
+        cfg.vector_store.embedding_dims = dims
+        cfg.history.db_path = str(tmp_path / "history.sqlite3")
+        store = ChromaVectorStore(cfg)
+        await store.initialize()
+        await store._get_entity_collection()
+        coding = []
+        for variant in ("working", "coding"):
+            cs = CodingMemoryStore(cfg, FakeEmbedder("m", dims or 8), variant=variant)
+            await cs.initialize()
+            coding.append(cs)
+        opened = {c.name for c in _client(tmp_path / "v").list_collections()}
+        names = vector_space_names(cfg)
+        assert set(names) == set(VECTOR_ROLES)
+        assert set(names.values()) == opened
+        suffix = f"_{dims}" if dims else ""
+        assert names["memory"] == f"memhub_abc123{suffix}"
+        assert names["coding_keys_cv"] == f"memhub_abc123_coding_keys_cv{suffix}"
+        await store.close()
+        for cs in coding:
+            await cs.close()
+
+    def test_explicit_base_name_and_space(self, tmp_path):
+        cfg = MemoryConfig()
+        cfg.vector_store.provider = "chroma"
+        cfg.vector_store.embedding_dims = 1024
+        names = vector_space_names(cfg, base_name="staging_x")
+        assert names["memory"] == "staging_x_1024"
+        assert names["entities"] == "staging_x_1024_entities"
+        assert names["working_facts"] == "staging_x_working_facts_1024"
+        assert vector_space_names(_space(tmp_path, "b", 512))["memory"] == "b_512"
+
+    def test_parse_collection_name_round_trips(self):
+        for dims in (512, None):
+            for role, name in vector_space_names(VectorSpace("memhub_9f", dims)).items():
+                assert parse_collection_name(name) == ("memhub_9f", dims, role)
+        assert parse_collection_name("memhub_9f_512_tag_index") is None
+
+    def test_space_identity_is_model_and_width_only(self):
+        a = space_identity("BAAI/bge-small-zh-v1.5", 512)
+        b = space_identity(" BAAI/bge-small-zh-v1.5 ", 512)
+        assert a == b and a.key == "BAAI/bge-small-zh-v1.5#512"
+        assert a.digest != space_identity("BAAI/bge-small-zh-v1.5", 1024).digest
+        with pytest.raises(ValueError):
+            space_identity("", 512)
+
+    def test_non_chroma_provider_is_not_supported(self):
+        cfg = MemoryConfig()
+        cfg.vector_store.provider = "qdrant"
+        with pytest.raises(NotImplementedError):
+            vector_space_names(cfg)
+
+
+class TestReembedText:
+    def test_canonical_state_uses_state_retrieval_text(self):
+        meta = {"content": "住在柏林", "custom": json.dumps(_custom_state())}
+        assert reembed_text("memory", "住在柏林", meta) == "user lives in Berlin (home city)"
+
+    def test_retrieval_text_when_present(self):
+        meta = {"custom": json.dumps({"retrieval_text": "  pattern: plans ahead  "})}
+        assert reembed_text("memory", "画像", meta) == "pattern: plans ahead"
+
+    def test_document_otherwise_with_content_fallback(self):
+        assert reembed_text("memory", "plain fact", {"custom": "{}"}) == "plain fact"
+        assert reembed_text("memory", None, {"content": "from metadata"}) == "from metadata"
+        assert reembed_text("memory", "doc", {"custom": "not json"}) == "doc"
+
+    def test_coding_key_and_working_fact_use_the_document(self):
+        coding = {"custom": json.dumps({"coding_key_text": "fix flaky test",
+                                        "retrieval_text": "ignored outside memory"})}
+        assert reembed_text("coding_keys", "fix flaky test", coding) == "fix flaky test"
+        assert reembed_text("working_facts", "repo: uses uv", {}) == "repo: uses uv"
+
+
+# ----------------------------------------------------------------------
+# re-embed
+# ----------------------------------------------------------------------
+
+@pytest.mark.parametrize(
+    "src_dims,tgt_dims,src_model,tgt_model",
+    [(512, 512, "old-model", "new-model"), (512, 1024, "old", "new"), (1024, 512, "old", "new")],
+)
+async def test_reembed_copies_everything_and_never_touches_source(
+    tmp_path, src_dims, tgt_dims, src_model, tgt_model,
+):
+    path = tmp_path / "v"
+    names = await _seed(path, "memhub_src", src_dims, model=src_model)
+    before = _snapshot(path, names)
+    source = _space(path, "memhub_src", src_dims)
+    target = _space(path, "memhub_dst", tgt_dims, model=tgt_model)
+    new = FakeEmbedder(tgt_model, tgt_dims)
+
+    plan = await plan_reembed(source, target)
+    assert plan.per_role["memory"] == 14 and plan.per_role["entities"] == 2
+    assert plan.total_vectors == 18 == plan.estimated_vectors
+    assert plan.source_dims == src_dims and plan.target_dims == tgt_dims
+
+    seen = []
+    report = await reembed_vector_space(
+        source, target, embedding=new, batch_size=4, on_progress=seen.append,
+    )
+    assert report.embedded == 18 and report.reused == 0
+    assert seen and seen[-1].done == seen[-1].total
+
+    assert _snapshot(path, names) == before  # source unchanged, byte for byte
+    target_names = vector_space_names(target)
+    for role in ("memory", "entities", "coding_keys", "working_facts_cv"):
+        src = _records(path, names[role])
+        dst = _records(path, target_names[role])
+        assert set(src) == set(dst)
+        for rid, (doc, meta, _) in src.items():
+            assert dst[rid][0] == doc
+            assert dst[rid][1] == meta  # status, is_latest, superseded_by, timestamps, custom
+            assert len(dst[rid][2]) == tgt_dims
+            assert dst[rid][2] == pytest.approx(new.vector(reembed_text(role, doc, meta)), abs=1e-5)
+    memory = _records(path, target_names["memory"])
+    layers = {meta["layer"] for _, meta, _ in memory.values()}
+    users = {meta["user_id"] for _, meta, _ in memory.values()}
+    assert "l1_raw" in layers and users == {"alice", "bob"}
+    statuses = {meta["status"] for _, meta, _ in memory.values()}
+    assert statuses == {"active", "superseded"}
+    state = [r for r in memory.values() if "canonical_state" in r[1]["custom"]]
+    assert state and state[0][2] == pytest.approx(
+        new.vector("user lives in Berlin (home city)"), abs=1e-5,
+    )
+    meta = _client(path).get_collection(target_names["memory"]).metadata
+    assert meta["hnsw:space"] == "cosine" and meta["hy_embedding_model"] == tgt_model
+
+    verified = await verify_reembed(source, target, embedding=new)
+    assert verified.ok, verified.problems
+    assert verified.per_user == {"alice": 10, "bob": 8}
+    assert verified.self_recall_samples == 18 and verified.self_recall_hits == 18
+
+
+async def test_resume_after_kill_only_embeds_the_rest(tmp_path):
+    path = tmp_path / "v"
+    await _seed(path, "src", 512, per_user=20)
+    source, target = _space(path, "src", 512), _space(path, "dst", 1024)
+
+    class Killed(Exception):
+        pass
+
+    first = FakeEmbedder("new", 1024)
+    batches = {"n": 0}
+
+    def _kill():
+        batches["n"] += 1
+        if batches["n"] > 3:
+            raise Killed()
+        return False
+
+    with pytest.raises(Killed):
+        await reembed_vector_space(
+            source, target, embedding=first, batch_size=5, max_inflight=1,
+            roles=["memory"], cancel=_kill,
+        )
+    assert first.texts == 15
+    second = FakeEmbedder("new", 1024)
+    report = await reembed_vector_space(
+        source, target, embedding=second, batch_size=5, max_inflight=1, roles=["memory"],
+    )
+    assert second.texts == 40 - 15
+    assert report.per_role["memory"].reused == 15
+    # The end state equals an uninterrupted run: every source record, verbatim,
+    # with the vector of its own text.
+    resumed = _records(path, "dst_1024")
+    src = _records(path, "src_512")
+    assert set(resumed) == set(src)
+    for rid, (doc, meta, vector) in resumed.items():
+        assert (doc, meta) == src[rid][:2]
+        assert vector == pytest.approx(second.vector(reembed_text("memory", doc, meta)), abs=1e-5)
+    assert (await verify_reembed(source, target, roles=["memory"], samples=0)).ok
+
+
+async def test_cancel_at_batch_boundary_leaves_no_partial_batch(tmp_path):
+    path = tmp_path / "v"
+    await _seed(path, "src", 512)
+    calls = {"n": 0}
+
+    def _cancel():
+        calls["n"] += 1
+        return calls["n"] > 2
+
+    with pytest.raises(ReembedCancelled) as info:
+        await reembed_vector_space(
+            _space(path, "src", 512), _space(path, "dst", 512),
+            embedding=FakeEmbedder("new", 512), batch_size=3, max_inflight=1,
+            roles=["memory"], cancel=_cancel,
+        )
+    assert info.value.report.per_role["memory"].embedded == 6
+    assert _client(path).get_collection("dst_512").count() == 6
+
+
+async def test_wrong_width_raises_before_any_upsert(tmp_path):
+    path = tmp_path / "v"
+    await _seed(path, "src", 512)
+    bad = FakeEmbedder("new", 1024, wrong_width_after=2)
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        await reembed_vector_space(
+            _space(path, "src", 512), _space(path, "dst", 1024),
+            embedding=bad, batch_size=4, roles=["memory"],
+        )
+    assert _client(path).get_collection("dst_1024").count() == 0
+    declared_wrong = FakeEmbedder("new", 768)
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        await reembed_vector_space(
+            _space(path, "src", 512), _space(path, "dst", 1024), embedding=declared_wrong,
+        )
+
+
+async def test_refuses_to_write_into_the_source_collection(tmp_path):
+    path = tmp_path / "v"
+    await _seed(path, "src", 512)
+    with pytest.raises(ValueError):
+        await reembed_vector_space(
+            _space(path, "src", 512), _space(path, "src", 512),
+            embedding=FakeEmbedder("new", 512),
+        )
+
+
+async def test_verify_detects_missing_extra_drift_and_width(tmp_path):
+    path = tmp_path / "v"
+    await _seed(path, "src", 512)
+    source, target = _space(path, "src", 512), _space(path, "dst", 512)
+    new = FakeEmbedder("new", 512)
+    await reembed_vector_space(source, target, embedding=new, roles=["memory"])
+    coll = _client(path).get_collection("dst_512")
+    got = coll.get(include=["documents", "metadatas"])
+    victim, drifted = got["ids"][0], got["ids"][1]
+    coll.delete(ids=[victim])
+    meta = dict(got["metadatas"][1])
+    meta["status"] = "archived"
+    coll.update(ids=[drifted], metadatas=[meta])
+    coll.add(ids=["stranger"], embeddings=[new.vector("x")], documents=["x"], metadatas=[{"user_id": "eve"}])
+    result = await verify_reembed(source, target, embedding=new, roles=["memory"], samples=5)
+    assert not result.ok
+    counts = result.per_role["memory"]
+    assert (counts["missing"], counts["extra"], counts["mismatched"]) == (1, 1, 1)
+    assert any("eve" in p for p in result.problems)
+
+    wide = _space(path, "dst", 1024)
+    wide_coll = _client(path).get_or_create_collection("dst_1024", metadata={"hnsw:space": "cosine"})
+    src = _records(path, "src_512")
+    wide_coll.add(
+        ids=list(src),
+        embeddings=[[0.0] * 512 for _ in src],
+        documents=[v[0] for v in src.values()],
+        metadatas=[v[1] for v in src.values()],
+    )
+    result = await verify_reembed(source, wide, roles=["memory"], samples=0)
+    assert result.per_role["memory"]["bad_dims"] == len(src) and not result.ok
+
+
+async def test_drop_vector_space_refuses_live_collections(tmp_path):
+    from hy_memory.data.vector_store_chroma import ChromaVectorStore
+
+    path = tmp_path / "v"
+    await _seed(path, "src", 512)
+    cfg = MemoryConfig()
+    cfg.vector_store.provider = "chroma"
+    cfg.vector_store.persist_directory = str(path)
+    cfg.vector_store.collection_name = "src"
+    cfg.vector_store.embedding_dims = 512
+    live = ChromaVectorStore(cfg)
+    await live.initialize()
+    with pytest.raises(RuntimeError):
+        await drop_vector_space(_space(path, "src", 512))
+    await live.close()
+    dropped = await drop_vector_space(_space(path, "src", 512))
+    assert set(dropped) == {"src_512", "src_512_entities", "src_coding_keys_512", "src_working_facts_cv_512"}
+    assert not {c.name for c in _client(path).list_collections()}
+
+
+# ----------------------------------------------------------------------
+# scan, diff, merge
+# ----------------------------------------------------------------------
+
+async def _two_spaces(path):
+    """A live 1024 space and a stray 512 space sharing some ids."""
+    from hy_memory.data.vector_store_chroma import ChromaVectorStore
+
+    async def _store(base, dims, model, contents):
+        cfg = MemoryConfig()
+        cfg.vector_store.provider = "chroma"
+        cfg.vector_store.persist_directory = str(path)
+        cfg.vector_store.collection_name = base
+        cfg.vector_store.embedding_dims = dims
+        store = ChromaVectorStore(cfg)
+        await store.initialize()
+        emb = FakeEmbedder(model, dims)
+        nodes = []
+        for node_id, user, content, status in contents:
+            node = MemoryNode(
+                node_id=node_id, user_id=user, agent_id="a", session_id="s",
+                layer=MemoryLayer.L2_FACT, content=content, status=status,
+            )
+            node.embedding = emb.vector(content)
+            nodes.append(node)
+        await store.upsert_batch(nodes)
+        await store.close()
+
+    ids = [f"00000000-0000-0000-0000-00000000000{i}" for i in range(6)]
+    await _store("memhub_old", 512, "bge", [
+        (ids[0], "alice", "shared one", MemoryStatus.ACTIVE),
+        (ids[1], "alice", "shared two (stray wording)", MemoryStatus.ACTIVE),
+        (ids[2], "bob", "only in stray", MemoryStatus.ACTIVE),
+        (ids[3], "bob", "only in stray, deleted", MemoryStatus.SHADOW),
+    ])
+    await _store("memhub_new", 1024, "e5", [
+        (ids[0], "alice", "shared one", MemoryStatus.ACTIVE),
+        (ids[1], "alice", "shared two (live wording)", MemoryStatus.ARCHIVED),
+        (ids[4], "carol", "only in live", MemoryStatus.ACTIVE),
+    ])
+    return ids
+
+
+async def test_scan_diff_and_merge(tmp_path):
+    path = tmp_path / "v"
+    ids = await _two_spaces(path)
+    spaces = await scan_vector_spaces(path)
+    by_base = {(s.base_name, s.dims): s for s in spaces}
+    assert set(by_base) == {("memhub_old", 512), ("memhub_new", 1024)}
+    old, live = by_base[("memhub_old", 512)], by_base[("memhub_new", 1024)]
+    assert old.roles["memory"].count == 4 and old.width == 512
+    assert live.roles["memory"].count == 3 and live.width == 1024
+    assert old.per_user == {"alice": 2, "bob": 2}
+    assert live.per_user == {"alice": 2, "carol": 1}
+
+    stray, target = old.space(), _space(path, "memhub_new", 1024, model="e5")
+    diff = await diff_spaces(stray, target)
+    assert set(diff.missing["memory"]) == {ids[2], ids[3]}  # status ignored
+    assert diff.total_missing == 2
+
+    live_before = _records(path, "memhub_new_1024")
+    stray_before = _snapshot(path, stray.names())
+    merger = FakeEmbedder("e5", 1024)
+    report = await reembed_vector_space(stray, target, embedding=merger)
+    assert merger.texts == 2 and report.embedded == 2
+    assert report.reused == 1 and report.kept == 1
+    after = _records(path, "memhub_new_1024")
+    assert set(after) == set(live_before) | {ids[2], ids[3]}
+    for rid, record in live_before.items():  # the target wins
+        assert after[rid][:2] == record[:2]
+        assert after[rid][2] == pytest.approx(record[2], abs=1e-6)
+    assert _snapshot(path, stray.names()) == stray_before
+    assert (await diff_spaces(stray, target)).empty
+
+    again = FakeEmbedder("e5", 1024)
+    second = await reembed_vector_space(stray, target, embedding=again)
+    assert again.texts == 0 and second.embedded == 0
+    assert _records(path, "memhub_new_1024").keys() == after.keys()
+    rescanned = {(s.base_name, s.dims): s for s in await scan_vector_spaces(path)}
+    assert rescanned[("memhub_new", 1024)].roles["memory"].count == 5
+
+
+async def test_scan_of_missing_directory_is_empty(tmp_path):
+    assert await scan_vector_spaces(tmp_path / "nope") == []
+
+
+# ----------------------------------------------------------------------
+# EmbedService fixes
+# ----------------------------------------------------------------------
+
+def _fake_raw(n, dims):
+    items = []
+    for i in range(n):
+        item = MagicMock()
+        item.embedding = [0.1] * dims
+        item.index = i
+        items.append(item)
+    resp = MagicMock()
+    resp.model = "m"
+    resp.usage = None
+    resp.data = items
+    raw = MagicMock()
+    raw.status_code = 200
+    raw.headers = {}
+    raw.parse.return_value = resp
+    return raw
+
+
+def _service(dims=8, **embedder):
+    cfg = MemoryConfig(embedder=EmbedderConfig(
+        provider="openai", model="m", api_key="sk", base_url="https://e.invalid/v1",
+        embedding_dims=dims, max_retries=2, retry_delay=0.0, **embedder,
+    ))
+    return EmbedService(cfg)
+
+
+async def test_embed_batch_chunks_to_batch_size():
+    svc = _service()
+    sizes = []
+
+    async def _create(**kwargs):
+        sizes.append(len(kwargs["input"]))
+        return _fake_raw(len(kwargs["input"]), 8)
+
+    client = MagicMock()
+    client.embeddings.with_raw_response.create = AsyncMock(side_effect=_create)
+    with patch("openai.AsyncOpenAI", return_value=client):
+        vectors = await svc.embed_batch([f"t{i}" for i in range(1000)], use_cache=False)
+    assert len(vectors) == 1000
+    assert len(sizes) == math.ceil(1000 / 64) and max(sizes) == 64
+
+    small = _service(batch_size=10)
+    sizes.clear()
+    with patch("openai.AsyncOpenAI", return_value=client):
+        _, metas = await small.embed_batch_with_meta([f"u{i}" for i in range(25)], use_cache=False)
+    assert sizes == [10, 10, 5]
+    assert [m["batch_index"] for m in metas] == list(range(25))
+
+
+async def test_429_honours_retry_after():
+    svc = _service()
+
+    class RateLimited(Exception):
+        status_code = 429
+
+        def __init__(self):
+            super().__init__("rate limited")
+            self.response = MagicMock()
+            self.response.headers = {"retry-after": "2"}
+
+    client = MagicMock()
+    client.embeddings.with_raw_response.create = AsyncMock(
+        side_effect=[RateLimited(), _fake_raw(1, 8)],
+    )
+    sleeps = []
+
+    async def _sleep(seconds):
+        sleeps.append(seconds)
+
+    with patch("openai.AsyncOpenAI", return_value=client), \
+            patch("hy_memory.core.embed_service.asyncio.sleep", _sleep):
+        vectors = await svc.embed_batch(["x"], use_cache=False)
+    assert len(vectors) == 1
+    assert len(sleeps) == 1 and 2.0 <= sleeps[0] <= 2.5
+
+
+async def test_per_call_max_retries():
+    svc = _service()
+
+    class Down(Exception):
+        status_code = 503
+
+    client = MagicMock()
+    client.embeddings.with_raw_response.create = AsyncMock(side_effect=Down("down"))
+    with patch("openai.AsyncOpenAI", return_value=client), pytest.raises(Down):
+        await svc.embed_batch(["x"], use_cache=False, max_retries=5)
+    assert client.embeddings.with_raw_response.create.await_count == 6
+
+
+async def test_returned_width_is_validated_and_not_retried():
+    svc = _service(dims=8)
+    client = MagicMock()
+    client.embeddings.with_raw_response.create = AsyncMock(return_value=_fake_raw(2, 7))
+    with patch("openai.AsyncOpenAI", return_value=client), \
+            pytest.raises(EmbeddingDimensionMismatchError):
+        await svc.embed_batch(["a", "b"], use_cache=False)
+    assert client.embeddings.with_raw_response.create.await_count == 1

tests/test_reembed_graph.py

@@ -0,0 +1,252 @@
+"""Re-embed the Kuzu graph into a new embedding space.
+
+The source graph is opened read-only and never modified. A new graph is built
+at the target path with the target width: every node and edge is copied, the
+embedding columns are recomputed from their text, the HNSW index is rebuilt.
+Runs in CI: kuzu is a core dependency.
+"""
+
+import hashlib
+import json
+import os
+
+import pytest
+
+from hy_memory import (
+    EmbeddingDimensionMismatchError,
+    MemoryConfig,
+    ReembedCancelled,
+    drop_graph,
+    reembed_graph,
+    verify_graph_reembed,
+)
+from hy_memory.models.memory import MemoryLayer, MemoryNode
+from hy_memory.reembed import hierarchy_topic_label, plan_graph_reembed
+
+from .test_reembed import FakeEmbedder
+
+
+def _cfg(path, dims):
+    cfg = MemoryConfig()
+    cfg.graph_store.provider = "kuzu"
+    cfg.graph_store.db_path = str(path)
+    cfg.vector_store.embedding_dims = dims
+    return cfg
+
+
+async def _open(path, dims):
+    from hy_memory.data.graph_store_kuzu import KuzuGraphStore
+
+    store = KuzuGraphStore(_cfg(path, dims))
+    await store.initialize()
+    assert store._available
+    return store
+
+
+def _node(content, *, layer=MemoryLayer.L2_FACT, custom=None, user="alice"):
+    return MemoryNode(
+        user_id=user, agent_id="a", session_id="s1",
+        layer=layer, content=content, custom=custom or {},
+    )
+
+
+async def _seed_graph(path, dims, model="old"):
+    """A graph with every node and edge type the SDK writes."""
+    old = FakeEmbedder(model, dims)
+    gs = await _open(path, dims)
+    schema = _node(
+        "画像：喜欢提前规划", layer=MemoryLayer.L6_SCHEMA,
+        custom={"schema_type": "basic", "retrieval_text": "plans trips weeks ahead",
+                "_has_beh": True, "beh_abstraction": "prepares early under uncertainty"},
+    )
+    schema._graph_embedding = old.vector("plans trips weeks ahead")
+    core = _node("core: anticipates", layer=MemoryLayer.L6_SCHEMA, custom={"schema_type": "abstract"})
+    core._graph_embedding = old.vector("core: anticipates")
+    stale_beh = _node("beh without text", layer=MemoryLayer.L6_SCHEMA, custom={})
+    stale_beh._graph_embedding = old.vector("beh without text")
+    facts = [_node(f"fact {i}", user="bob" if i % 2 else "alice") for i in range(5)]
+    for node in [schema, core, stale_beh, *facts]:
+        await gs.upsert_memory_node(node)
+    await gs.update_embedding(schema.node_id, beh_embedding=old.vector("prepares early under uncertainty"))
+    await gs.update_embedding(stale_beh.node_id, beh_embedding=old.vector("lost text"))
+    await gs.add_cross_abstracts_to(schema.node_id, core.node_id)
+    ik = facts[0].get_isolation_key()
+    await gs.add_edge(facts[0].node_id, facts[1].node_id, "RELATED_TO", {"relation_type": "co_occurs", "weight": 0.7})
+    await gs.add_temporal_edge(facts[0].node_id, facts[2].node_id, topic_id="t")
+    for fact in facts[:3]:
+        await gs.add_topic_tag(fact.node_id, "travel", ik, embed_service=old)
+    await gs.add_topic_tag(facts[3].node_id, "work", ik)  # no embed service: NULL embedding
+    await gs.add_topic_path(
+        facts[0].node_id, ["Travel", "Japan trip"], ik, "alice",
+        node_types=["topic", "series"], embed_service=old,
+    )
+    await gs.add_topic_path(facts[1].node_id, ["Work", "Standups"], ik, "alice", embed_service=old)
+    await gs.add_entity_mention(facts[0].node_id, "Kyoto", "GPE", ik)
+    await gs.ensure_vdbref("raw-1", "l1_raw")
+    await gs.add_derived_from(schema.node_id, "raw-1")
+    await gs.close()
+    return old, {"schema": schema, "core": core, "stale_beh": stale_beh, "facts": facts}
+
+
+def _digest(path):
+    """Hash of the Kuzu DB file and any sidecar files next to it."""
+    h = hashlib.sha256()
+    for name in sorted(os.listdir(path.parent)):
+        if name == path.name or name.startswith(path.name + "."):
+            h.update(name.encode())
+            with open(path.parent / name, "rb") as handle:
+                h.update(handle.read())
+    return h.hexdigest()
+
+
+def _read(path, query):
+    import kuzu
+
+    db = kuzu.Database(str(path), read_only=True)
+    conn = kuzu.Connection(db)
+    try:
+        result = conn.execute(query)
+        rows = []
+        while result.has_next():
+            rows.append(result.get_next())
+        return rows
+    finally:
+        conn.close()
+        db.close()
+
+
+@pytest.mark.parametrize(
+    "src_dims,tgt_dims,src_model,tgt_model",
+    [(512, 512, "old", "new"), (512, 1024, "old", "new"), (1024, 512, "old", "new")],
+)
+async def test_reembed_graph_copies_everything_at_the_new_width(
+    tmp_path, src_dims, tgt_dims, src_model, tgt_model,
+):
+    source, target = tmp_path / "graph", tmp_path / "graph-new"
+    _, nodes = await _seed_graph(source, src_dims, model=src_model)
+    before = _digest(source)
+    plan = await plan_graph_reembed(source, target_dims=tgt_dims)
+    assert plan.source_width == src_dims and plan.target_width == tgt_dims
+    assert plan.memory_nodes == 8 and plan.topics == 2 and plan.hierarchy_topics == 4
+    assert plan.memory_embeddings == 3 and plan.beh_embeddings == 2
+    assert plan.topic_embeddings == 1 and plan.hierarchy_topic_embeddings == 4
+
+    new = FakeEmbedder(tgt_model, tgt_dims)
+    seen = []
+    report = await reembed_graph(
+        source, target, embedding=new, target_dims=tgt_dims, batch_size=3, on_progress=seen.append,
+    )
+    assert _digest(source) == before  # the source graph is untouched
+    assert not os.path.exists(str(target) + ".reembed-partial")
+    assert report.source_width == src_dims and report.target_width == tgt_dims
+    assert report.node_counts == plan.node_counts
+    assert report.rel_counts == plan.rel_counts
+    assert report.embedded == {
+        "Memory.embedding": 3, "Memory.beh_embedding": 1,
+        "Topic.embedding": 1, "HierarchyTopic.embedding": 4,
+    }
+    assert report.cleared == {"Memory.beh_embedding": 1}  # beh vector with no beh_abstraction
+    assert report.hierarchy_labels_exact == 4 and report.hierarchy_labels_inexact == ()
+    assert report.vector_indexes == ("memory_content_idx",)
+    assert seen
+
+    check = await verify_graph_reembed(source, target, target_dims=tgt_dims)
+    assert check.ok, check.problems
+    assert check.vector_index and check.target_width == tgt_dims
+    assert all(src == dst for src, dst in check.node_counts.values())
+    assert all(src == dst for src, dst in check.rel_counts.values())
+
+    schema = nodes["schema"]
+    rows = _read(target, f"MATCH (m:Memory {{node_id: '{schema.node_id}'}}) RETURN m.embedding, m.beh_embedding, m.custom_json, m.content;")
+    emb, beh, custom, content = rows[0]
+    assert len(emb) == tgt_dims
+    assert emb == pytest.approx(new.vector("plans trips weeks ahead"), abs=1e-5)
+    assert beh == pytest.approx(new.vector("prepares early under uncertainty"), abs=1e-5)
+    assert json.loads(custom)["retrieval_text"] == "plans trips weeks ahead" and content == schema.content
+    rows = _read(target, f"MATCH (m:Memory {{node_id: '{nodes['core'].node_id}'}}) RETURN m.embedding;")
+    assert rows[0][0] == pytest.approx(new.vector("core: anticipates"), abs=1e-5)
+    rows = _read(target, f"MATCH (m:Memory {{node_id: '{nodes['facts'][0].node_id}'}}) RETURN m.embedding IS NULL;")
+    assert rows[0][0] is True  # a NULL source vector stays NULL
+    topics = dict(_read(target, "MATCH (t:Topic) RETURN t.name, t.embedding;"))
+    assert topics["travel"] == pytest.approx(new.vector("travel"), abs=1e-5)
+    assert topics["work"] is None
+    for subject, path_json, embedding in _read(target, "MATCH (t:HierarchyTopic) RETURN t.subject, t.path_json, t.embedding;"):
+        label, exact = hierarchy_topic_label(subject, path_json)
+        assert exact and embedding == pytest.approx(new.vector(label), abs=1e-5)
+
+    # The rebuilt graph serves the SDK: vector search on the new index works.
+    gs = await _open(target, tgt_dims)
+    try:
+        hits = await gs.vector_search(
+            query_embedding=new.vector("plans trips weeks ahead"),
+            isolation_key=schema.get_isolation_key(), limit=1,
+        )
+        assert hits and hits[0]["node_id"] == schema.node_id
+        width = await gs._run("CALL table_info('Memory') RETURN *;")
+        assert dict((r[1], r[2]) for r in width)["embedding"] == f"FLOAT[{tgt_dims}]"
+    finally:
+        await gs.close()
+
+
+async def test_source_labels_rebuild_exactly(tmp_path):
+    """The stored HierarchyTopic vectors were computed from the rebuilt label."""
+    source = tmp_path / "graph"
+    old, _ = await _seed_graph(source, 64)
+    rows = _read(source, "MATCH (t:HierarchyTopic) RETURN t.subject, t.path_json, t.embedding;")
+    assert len(rows) == 4
+    for subject, path_json, embedding in rows:
+        label, exact = hierarchy_topic_label(subject, path_json)
+        assert exact
+        assert embedding == pytest.approx(old.vector(label), abs=1e-6)
+
+
+def test_legacy_path_json_is_flagged_inexact():
+    label, exact = hierarchy_topic_label("alice", json.dumps(["Travel", "Japan"]), "Japan")
+    assert label == "alice: topic:Travel > facet:Japan" and not exact
+    label, exact = hierarchy_topic_label("alice", "", "Japan")
+    assert label == "alice: Japan" and not exact
+
+
+async def test_cancel_and_failure_leave_no_target(tmp_path):
+    source, target = tmp_path / "graph", tmp_path / "graph-new"
+    await _seed_graph(source, 64)
+    with pytest.raises(ReembedCancelled):
+        await reembed_graph(
+            source, target, embedding=FakeEmbedder("new", 32), target_dims=32,
+            cancel=lambda: True,
+        )
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        await reembed_graph(
+            source, target, embedding=FakeEmbedder("new", 32, wrong_width_after=1), target_dims=32,
+        )
+    assert not os.path.exists(target)
+    assert not os.path.exists(str(target) + ".reembed-partial")
+    # A restart starts over and succeeds.
+    report = await reembed_graph(source, target, embedding=FakeEmbedder("new", 32), target_dims=32)
+    assert report.node_counts["Memory"] == 8
+    with pytest.raises(FileExistsError):
+        await reembed_graph(source, target, embedding=FakeEmbedder("new", 32), target_dims=32)
+    again = await reembed_graph(
+        source, target, embedding=FakeEmbedder("new", 32), target_dims=32, overwrite=True,
+    )
+    assert again.rel_counts == report.rel_counts
+
+
+async def test_refuses_open_source_and_drops_only_closed_graphs(tmp_path):
+    source, target = tmp_path / "graph", tmp_path / "graph-new"
+    await _seed_graph(source, 64)
+    live = await _open(source, 64)
+    try:
+        with pytest.raises(RuntimeError):
+            await reembed_graph(source, target, embedding=FakeEmbedder("new", 64), target_dims=64)
+        with pytest.raises(RuntimeError):
+            await drop_graph(source)
+    finally:
+        await live.close()
+    with pytest.raises(ValueError):
+        await reembed_graph(source, source, embedding=FakeEmbedder("new", 64), target_dims=64)
+    await reembed_graph(source, target, embedding=FakeEmbedder("new", 64), target_dims=64)
+    assert await drop_graph(target) is True
+    assert not os.path.exists(target)
+    assert await drop_graph(target) is False
+    assert os.path.exists(source)