Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(memory): keep vector metadata in sync (#2295)
Chesterc471800e8ee2
diff --git a/headroom/memory/backends/local.py b/headroom/memory/backends/local.pyindex 6fafeb91b..98e94b7e6 100644--- a/headroom/memory/backends/local.py+++ b/headroom/memory/backends/local.py@@ -528,6 +528,12 @@ class LocalBackend:assert self._hierarchical_memory is not Nonereturn await self._hierarchical_memory.record_access(memory_ids)+ async def refresh_memory_indexes(self, memory_id: str) -> Memory | None:+ """Refresh vector metadata and cache state from the primary store."""+ await self._ensure_initialized()+ assert self._hierarchical_memory is not None+ return await self._hierarchical_memory.refresh_memory_indexes(memory_id)+async def update_memory(self,memory_id: str,diff --git a/headroom/memory/core.py b/headroom/memory/core.pyindex 0265b8e12..776d5152c 100644--- a/headroom/memory/core.py+++ b/headroom/memory/core.py@@ -494,10 +494,14 @@ class HierarchicalMemory:# Save updatesawait self._store.save(memory)- # Update indexes+ # Vector indexes also carry filterable metadata. Refresh them for+ # metadata-only updates, not just when the embedding content changes.+ vector_metadata_changed = any(+ value is not None for value in (importance, entity_refs, metadata)+ )+ if (content_changed or vector_metadata_changed) and memory.embedding is not None:+ await self._vector_index.index(memory)if content_changed:- if memory.embedding is not None:- await self._vector_index.index(memory)await self._index_for_text_search(memory)# Invalidate and re-cache@@ -507,6 +511,21 @@ class HierarchicalMemory:return memory+ async def refresh_memory_indexes(self, memory_id: str) -> Memory | None:+ """Refresh secondary index metadata from the primary memory store."""+ memory = await self._store.get(memory_id)+ if memory is None:+ return None++ if memory.embedding is not None:+ await self._vector_index.index(memory)++ if self._cache is not None:+ await self._cache.invalidate(memory_id)+ await self._cache.put(memory)++ return memory+async def supersede(self,old_memory_id: str,diff --git a/headroom/memory/traffic_learner.py b/headroom/memory/traffic_learner.pyindex abd200ff3..387813911 100644--- a/headroom/memory/traffic_learner.py+++ b/headroom/memory/traffic_learner.py@@ -1389,10 +1389,10 @@ class TrafficLearner:now_iso = datetime.now(timezone.utc).isoformat()- def _bump() -> None:+ def _bump() -> bool:conn = sqlite3.connect(str(db_path))try:- conn.execute(+ cursor = conn.execute("UPDATE memories SET metadata = json_set(""metadata, '$.evidence_count', ""COALESCE(json_extract(metadata, '$.evidence_count'), 0) + 1, "@@ -1401,13 +1401,26 @@ class TrafficLearner:(now_iso, memory_id),)conn.commit()+ return cursor.rowcount > 0finally:conn.close()try:- await asyncio.to_thread(_bump)+ updated = await asyncio.to_thread(_bump)except Exception as e:logger.debug("Traffic learner evidence bump failed for %s: %s", memory_id, e)+ return++ refresh = getattr(self._backend, "refresh_memory_indexes", None)+ if updated and refresh is not None:+ try:+ await refresh(memory_id)+ except Exception as e:+ logger.debug(+ "Traffic learner evidence index refresh failed for %s: %s",+ memory_id,+ e,+ )# =========================================================================# Convenience: Extract from Anthropic messages formatdiff --git a/tests/test_memory/test_core_metadata_index_sync.py b/tests/test_memory/test_core_metadata_index_sync.pynew file mode 100644index 000000000..c017f7ee4--- /dev/null+++ b/tests/test_memory/test_core_metadata_index_sync.py@@ -0,0 +1,59 @@+"""Regression tests for primary-store and vector-metadata consistency."""++from unittest.mock import AsyncMock++import numpy as np+import pytest++from headroom.memory.core import HierarchicalMemory+from headroom.memory.models import Memory+++def _memory_system(memory: Memory, *, cache=None):+ store = AsyncMock()+ store.get.return_value = memory+ vector_index = AsyncMock()+ text_index = AsyncMock()+ embedder = AsyncMock()+ system = HierarchicalMemory(store, vector_index, text_index, embedder, cache=cache)+ return system, store, vector_index, text_index+++async def test_metadata_only_update_refreshes_vector_metadata() -> None:+ memory = Memory(+ id="memory-1",+ content="stable content",+ user_id="default",+ embedding=np.array([0.1, 0.2], dtype=np.float32),+ metadata={"evidence_count": 5},+ )+ system, store, vector_index, text_index = _memory_system(memory)++ updated = await system.update(memory.id, metadata={"evidence_count": 7})++ assert updated is memory+ store.save.assert_awaited_once_with(memory)+ vector_index.index.assert_awaited_once_with(memory)+ text_index.index.assert_not_awaited()+++async def test_refresh_memory_indexes_reloads_store_and_cache() -> None:+ memory = Memory(+ id="memory-1",+ content="stable content",+ user_id="default",+ embedding=np.array([0.1, 0.2], dtype=np.float32),+ metadata={"evidence_count": 7},+ )+ cache = AsyncMock()+ system, store, vector_index, _ = _memory_system(memory, cache=cache)++ refreshed = await system.refresh_memory_indexes(memory.id)++ assert refreshed is memory+ store.get.assert_awaited_once_with(memory.id)+ vector_index.index.assert_awaited_once_with(memory)+ cache.invalidate.assert_awaited_once_with(memory.id)+ cache.put.assert_awaited_once_with(memory)diff --git a/tests/test_memory/test_traffic_learner.py b/tests/test_memory/test_traffic_learner.pyindex fae5bc8fe..4faa037b8 100644--- a/tests/test_memory/test_traffic_learner.py+++ b/tests/test_memory/test_traffic_learner.py@@ -904,6 +904,10 @@ class _FakeBackend:self._config = _types.SimpleNamespace(db_path=str(db_path))self._db_path = str(db_path)+ self.refreshed_ids = []++ async def refresh_memory_indexes(self, memory_id: str):+ self.refreshed_ids.append(memory_id)async def save_memory(self,@@ -1465,6 +1469,7 @@ class TestBumpEdgeCases:learner = TrafficLearner(backend=backend, min_evidence=1)await learner._bump_persisted_evidence("no-such-id")assert _read_traffic_rows(db) == []+ assert backend.refreshed_ids == []# =============================================================================@@ -2092,6 +2097,7 @@ class TestBumpPersistsLastSeenAt:# Should be parseable back.parsed = _parse_iso_timestamp(meta["last_seen_at"])assert parsed is not None+ assert backend.refreshed_ids == ["row-1"]class TestHydrateLegacyRow: