fix(mem0): scope prefetch search and result cache by user_id

[dizhaky]

Summary

• queue_prefetch() was ignoring its user_id argument and searching under self._user_id (set once at initialize() time). In shared gateway sessions where the cached AIAgent is reused across participants, this caused cross-user memory leakage.
• prefetch() returned cached results without checking which user they were fetched for, so a result queued for User A could be injected into User B's next prompt.

Changes

• queue_prefetch(): capture effective_user_id = user_id or self._user_id before spawning the background thread; pass {"user_id": effective_user_id} directly to client.search() instead of self._read_filters(); store result as (user_id, text) tuple.
• prefetch(): discard the cached result when its owner user_id does not match the current caller — prevents stale cross-user results from leaking into the wrong session.
• __init__: initialize _prefetch_result to None instead of "" to match the new tuple-or-None shape.

Test plan

• In a shared gateway session (thread_sessions_per_user=False), send a message as User A, then immediately as User B — confirm User B does not receive User A's prefetched memories
• Single-user CLI session still prefetches and injects context correctly
• Verify existing unit tests in tests/plugins/memory/test_mem0_v2.py pass

Addresses Codex PR#33 P1 review comment: Scope prefetched memory by current user.

🤖 Generated with Claude Code

https://claude.ai/code/session_01DvT8QMaMr4m9dQUAcVdypv

---

Generated by Claude Code

---

Note

Medium Risk
Fixes a privacy-sensitive cross-user leak in prompt injection; change is small and localized to prefetch only.

Overview
Fixes cross-user memory leakage when a shared gateway reuses one Mem0MemoryProvider across participants: background prefetch no longer ignores the per-turn user_id.

queue_prefetch now resolves effective_user_id from the call (fallback self._user_id), searches with filters={"user_id": effective_user_id} instead of instance-scoped _read_filters(), and stores the cache as (user_id, text).

prefetch only returns cached text when the tuple’s owner matches the current caller; otherwise it drops the cache and returns empty, so User A’s queued results cannot appear in User B’s prompt.

_prefetch_result is initialized to None to match the tuple-or-None cache shape.

^{Reviewed by Cursor Bugbot for commit 1c04ca5. Configure here.}

[github-actions]

🔎 Lint report: fix/mem0-prefetch-user-scoping vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 8649 on HEAD, 8649 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 4570 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Empty prefetch still injects header
  • Added a check for an empty cached text element in prefetch so a (user_id, "") tuple no longer returns a header-only Mem0 section.

Or push these changes by commenting:

@cursor push 3ede6df721

Preview (3ede6df721)

diff --git a/plugins/memory/mem0/__init__.py b/plugins/memory/mem0/__init__.py
--- a/plugins/memory/mem0/__init__.py
+++ b/plugins/memory/mem0/__init__.py
@@ -241,7 +241,7 @@
         with self._prefetch_lock:
             cached = self._prefetch_result
             self._prefetch_result = None
-        if not cached or cached[0] != effective_user_id:
+        if not cached or cached[0] != effective_user_id or not cached[1]:
             return ""
         return f"## Mem0 Memory\n{cached[1]}"

_{You can send follow-ups to the cloud agent here.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 1c04ca5. Configure here.}

[cursor]

Empty prefetch still injects header

Low Severity

When queue_prefetch gets Mem0 hits but every memory field is empty, the cache becomes a (user_id, "") tuple. prefetch treats any truthy tuple as valid and returns ## Mem0 Memory with no bullets. Previously an empty joined string made prefetch return nothing, so prefetch_all skipped injection.

 

^{Reviewed by Cursor Bugbot for commit 1c04ca5. Configure here.}

[dizhaky]

Fix applied in commit 3a3d3ce — added or not cached[1] to the guard in prefetch(). The empty-text tuple case now correctly returns "" rather than injecting a bare ## Mem0 Memory header with no bullets.

Reviewed and confirmed by Claude Code.

---

Generated by Claude Code

[dizhaky]

Applied the Cursor Bugbot fix for Empty prefetch still injects header (commit 3a3d3ce).

Analysis: The bug is valid. In queue_prefetch, when Mem0 returns results but all memory fields are empty/falsy, lines becomes an empty list and the cache is set to (user_id, ""). The prefetch method was checking if not cached or cached[0] != effective_user_id — but a (user_id, "") tuple is truthy, so it would return ## Mem0 Memory\n with no bullet points.

Fix applied:

-        if not cached or cached[0] != effective_user_id:
+        if not cached or cached[0] != effective_user_id or not cached[1]:
             return ""

This is low-risk, correct, and matches the Bugbot autofix exactly.

---

Generated by Claude Code