Say when a saved memory becomes searchable

README.md

@@ -93,7 +93,7 @@ about it.
 | Tool | Purpose |
 |---|---|
 | `search_memory` | Relevance search over stored memories (optionally with the user profile) |
-| `add_memory` | Store a durable fact or exchange; saves in the background by default (`wait=true` to block until it is searchable) |
+| `add_memory` | Store a durable fact or exchange; saves in the background by default (`wait=true` to block until extraction has finished) |
 | `get_profile` | The synthesized user profile (facts, traits, preferences) |
 | `list_memories` | Chronological, paginated browsing |
 | `forget_session` | Delete what this connection stored (memories + cases distilled from its trajectories); the profile and learned skills are kept |
@@ -107,6 +107,9 @@ permission prompt; `forget_session` is marked destructive.
 
 - Profile updates and case/skill distillation run in an offline pipeline and land seconds
   to minutes after the write.
+- Extraction finishing and the memory becoming searchable are two steps: memories saved in
+  the last minute may not be indexed yet, and `list_memories` shows new entries sooner than
+  `search_memory`.
 - A background save that fails is reported on the next tool result, so the model never
   silently believes something was remembered.
 - Trajectories need more than three tool-call rounds to pass the distillation quality

src/everos_mcp/conversation.py

@@ -108,7 +108,8 @@ def reply(self, text: str) -> str:
 
     async def store(self, messages: list[dict[str, Any]], session_id: str) -> str | None:
         """Write messages, then make sure extraction ran.
-        Returns the final status; "extracted" means searchable."""
+        Returns the final status; "extracted" means extraction finished, not that
+        the memory is searchable yet (the index catches up afterwards)."""
         async with self._lock:
             return await self._add_and_flush(messages, session_id)
 

src/everos_mcp/server.py

@@ -209,10 +209,11 @@ async def add_memory(
     reports in Chinese"), a habit, a decision, or a long-term goal. Pass what
     the user said as user_message and/or your reply as assistant_message.
 
-    By default it returns at once and extraction runs in the background;
-    the memory is searchable a few seconds later, and a failure is reported
-    on a later tool result. Set wait=true only when you must search for it
-    right away — it blocks until extraction finishes.
+    By default it returns at once and extraction runs in the background,
+    and a failure is reported on a later tool result. Set wait=true to block
+    until extraction has finished. Either way the search index catches up
+    afterwards: memories saved in the last minute may not be indexed yet, and
+    list_memories shows them sooner than search_memory.
     """
     conv = _conversation(ctx)
     if not user_message and not assistant_message:
@@ -242,7 +243,10 @@ async def add_memory(
     if wait or conv.ephemeral or conv.saturated:
         status = await conv.store(messages, s.session_id)
         if status == "extracted":
-            return conv.reply("Stored and extracted; the memory is now searchable.")
+            return conv.reply(
+                "Stored and extracted. Memories saved in the last minute may not be "
+                "indexed yet; list_memories shows them sooner."
+            )
         return conv.reply(
             f"Stored, but extraction did not run (status {status!r}). The memory is "
             "NOT searchable yet — do not tell the user it was remembered."

tests/test_everos_mcp.py

@@ -242,7 +242,8 @@ def test_add_memory_always_flushes_the_open_tail(tool_client):
         }
     )
     reply = run(server.add_memory(user_message="x", wait=True))
-    assert "now searchable" in reply
+    assert "Stored and extracted" in reply
+    assert "now searchable" not in reply
     assert paths(seen) == ["add", "flush"]
 
 
@@ -747,7 +748,8 @@ async def scenario():
         return saved, forgot
 
     saved, forgot = run(scenario())
-    assert "now searchable" in saved  # waited instead of backgrounding
+    assert "Stored and extracted" in saved  # waited instead of backgrounding
+    assert "now searchable" not in saved
     assert "Nothing to forget" in forgot
     assert [r.url.path.rsplit("/", 1)[-1] for r in seen] == ["add", "flush"]
 
@@ -838,7 +840,8 @@ def test_saturated_conversation_saves_in_the_foreground(tool_client):
     for _ in range(16):
         conv._pending.add(object())  # stand-ins for running saves
     reply = run(server.add_memory(user_message="x"))
-    assert "now searchable" in reply
+    assert "Stored and extracted" in reply
+    assert "now searchable" not in reply
 
 
 def test_notices_are_capped():