From c5fafe56bb5a4e36bfdbad6de50f3672169b3b79 Mon Sep 17 00:00:00 2001
From: Krrish Dholakia <krrish+github@berri.ai>
Date: Fri, 24 Apr 2026 09:08:53 -0700
Subject: [PATCH] fix(memory): compose visibility + key filters via explicit
 AND

Greptile P1 (filter-fragility): `where.update(vis)` was semantically
correct today, but dict-merging by key meant any future visibility
filter that grew a new top-level "OR" would silently clobber the
existing key filter.

Compose explicitly instead:

    where = {"AND": [key_filter, vis]}

Applied to both `list_memory` and `_find_memory_for_caller`. When
either side is empty (admin has no visibility filter; list has no
key filter), skip the wrapper and use the non-empty side directly
to keep the generated SQL clean.

Test fake's `_matches` now understands top-level `AND` too.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 litellm/proxy/memory/memory_endpoints.py      | 31 ++++++++++++-------
 .../proxy/memory/test_memory_endpoints.py     |  4 +++
 2 files changed, 23 insertions(+), 12 deletions(-)

diff --git a/litellm/proxy/memory/memory_endpoints.py b/litellm/proxy/memory/memory_endpoints.py
index 35e635bbc13..937647ce7a0 100644
--- a/litellm/proxy/memory/memory_endpoints.py
+++ b/litellm/proxy/memory/memory_endpoints.py
@@ -223,17 +223,25 @@ async def list_memory(
     """List memory entries visible to the caller."""
     prisma_client = _require_prisma()
 
-    where: dict = {}
-    vis = _visibility_filter(user_api_key_dict)
-    if vis is not None:
-        where.update(vis)
-    # key_prefix takes precedence over exact `key` if both are passed.
-    # Both sit under the visibility filter (ANDed), so prefix scans can
-    # never leak across user/team scopes.
+    # Build the key filter first (prefix wins if both `key` and `key_prefix`
+    # are passed). Then AND it with the visibility filter via an explicit
+    # top-level "AND" — safer than `dict.update` since future visibility
+    # filters could grow an "OR" key that would clobber this one if merged
+    # by key.
+    key_filter: dict = {}
     if key_prefix is not None:
-        where["key"] = {"startsWith": key_prefix}
+        key_filter["key"] = {"startsWith": key_prefix}
     elif key is not None:
-        where["key"] = key
+        key_filter["key"] = key
+
+    vis = _visibility_filter(user_api_key_dict)
+    where: dict
+    if vis is None:
+        where = key_filter
+    elif not key_filter:
+        where = vis
+    else:
+        where = {"AND": [key_filter, vis]}
 
     try:
         total = await prisma_client.db.litellm_memorytable.count(where=where)
@@ -254,10 +262,9 @@ async def _find_memory_for_caller(
     prisma_client: Any, key: str, user_api_key_dict: UserAPIKeyAuth
 ) -> Any:
     """Look up a memory row by key, scoped to the caller's visibility."""
-    where: dict = {"key": key}
+    key_filter = {"key": key}
     vis = _visibility_filter(user_api_key_dict)
-    if vis is not None:
-        where.update(vis)
+    where = key_filter if vis is None else {"AND": [key_filter, vis]}
     rows = await prisma_client.db.litellm_memorytable.find_many(
         where=where, take=1, order={"updated_at": "desc"}
     )
diff --git a/tests/test_litellm/proxy/memory/test_memory_endpoints.py b/tests/test_litellm/proxy/memory/test_memory_endpoints.py
index fe83327af1a..fd82ba9b391 100644
--- a/tests/test_litellm/proxy/memory/test_memory_endpoints.py
+++ b/tests/test_litellm/proxy/memory/test_memory_endpoints.py
@@ -55,6 +55,10 @@ class _InMemoryMemoryTable:
 
     def _matches(self, row: MagicMock, where: Dict[str, Any]) -> bool:
         for k, v in where.items():
+            if k == "AND":
+                if not all(self._matches(row, clause) for clause in v):
+                    return False
+                continue
             if k == "OR":
                 if not any(self._matches(row, clause) for clause in v):
                     return False