聊天和回忆录证据检索都走 pgvector，去掉 Postgres FTS/content_tsv，新迁移删掉 content_tsv 列（部署要先 alembic upgrade）。

Embedding 端口增加 is_available()，聊天和回忆录日志用统一方式表示向量是否真能调用。

记忆整理（compaction）支持 Beat 定期扫用户；

事实抽取提示与 subject 归一化，减少同一人多种称呼；

api/tests/test_memory_compaction.py

@@ -19,7 +19,10 @@ from app.features.memory.compaction_service import (
     text_layer_match,
 )
 from app.features.memory.service import ingest_transcript_sync
-from app.tasks.memory_compaction_tasks import memory_compaction_run
+from app.tasks.memory_compaction_tasks import (
+    memory_compaction_run,
+    memory_compaction_sweep,
+)
 
 
 class FakeRedis:
@@ -197,12 +200,17 @@ def test_ingest_transcript_sync_populates_embeddings(monkeypatch) -> None:
             self.commit_calls += 1
 
     class FakeEmbeddingProvider:
+        def is_available(self) -> bool:
+            return True
+
         async def embed_texts(self, texts: list[str]) -> list[list[float]]:
             return [[float(i)] for i, _ in enumerate(texts, start=1)]
 
+        def embed_texts_sync(self, texts: list[str]) -> list[list[float]]:
+            return [[float(i)] for i, _ in enumerate(texts, start=1)]
+
     fake_session = FakeSession()
     embedded: list[tuple[str, list[float]]] = []
-    fts_updated: list[str] = []
 
     monkeypatch.setattr(settings, "memory_enrichment_enabled", False)
     monkeypatch.setattr(
@@ -221,10 +229,6 @@ def test_ingest_transcript_sync_populates_embeddings(monkeypatch) -> None:
         "app.features.memory.repo.create_chunk_sync",
         lambda *args, **kwargs: SimpleNamespace(id=f"chunk-{kwargs['chunk_index']}"),
     )
-    monkeypatch.setattr(
-        "app.features.memory.repo.update_chunk_fts_sync",
-        lambda session, chunk_id: fts_updated.append(chunk_id),
-    )
     monkeypatch.setattr(
         "app.features.memory.repo.update_chunk_embedding_sync",
         lambda session, chunk_id, embedding: embedded.append((chunk_id, embedding)),
@@ -239,7 +243,6 @@ def test_ingest_transcript_sync_populates_embeddings(monkeypatch) -> None:
 
     assert source_id == "src-1"
     assert [chunk_id for chunk_id, _ in embedded] == ["chunk-0", "chunk-1"]
-    assert fts_updated == ["chunk-0", "chunk-1"]
     assert fake_session.commit_calls == 1
 
 
@@ -442,3 +445,53 @@ def test_memory_compaction_run_releases_gate_and_retries_on_failure(
     assert "retry:RuntimeError" in events
     assert "release_lock" in events
     assert events.index("release_gate") < events.index("retry:RuntimeError")
+
+
+def test_memory_compaction_sweep_skipped_when_disabled(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(settings, "memory_compaction_enabled", False)
+    out = memory_compaction_sweep()
+    assert out == {"skipped": True, "reason": "disabled"}
+
+
+def test_memory_compaction_sweep_schedules_recent_users(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(settings, "memory_compaction_enabled", True)
+    monkeypatch.setattr(settings, "memory_compaction_sweep_recent_hours", 24)
+    scheduled: list[tuple[str, dict]] = []
+
+    class _DbCtx:
+        def __enter__(self):
+            return object()
+
+        def __exit__(self, *args):
+            return None
+
+    monkeypatch.setattr(
+        "app.tasks.memory_compaction_tasks.get_sync_db",
+        lambda: _DbCtx(),
+    )
+
+    def fake_list(session, *, hours):
+        assert hours == 24
+        return ["user-a", "user-b"]
+
+    monkeypatch.setattr(
+        "app.tasks.memory_compaction_tasks.list_users_with_recent_chunks_sync",
+        fake_list,
+    )
+
+    monkeypatch.setattr(
+        "app.tasks.memory_compaction_tasks.schedule_memory_compaction_run",
+        lambda uid, ctx: scheduled.append((uid, dict(ctx))),
+    )
+
+    out = memory_compaction_sweep()
+    assert out["scheduled"] == 2
+    assert set(out["user_ids"]) == {"user-a", "user-b"}
+    assert {u for u, _ in scheduled} == {"user-a", "user-b"}
+    for _, ctx in scheduled:
+        assert ctx.get("trigger_source") == "beat"
+        assert ctx.get("sweep_hours") == 24

api/tests/test_memory_evidence.py

@@ -1,14 +1,69 @@
 """Memory evidence 组装与检索契约（纯函数 / 无 DB）。"""
 
+import pytest
+
+from app.features.memory import evidence as evidence_mod
 from app.features.memory.evidence import (
     EMPTY_EVIDENCE_BUNDLE,
     _facts_to_dicts,
     _stories_to_dicts,
     _timeline_to_dicts,
+    retrieve_evidence_bundle_sync,
 )
 from app.features.memory.schemas import EvidenceBundle
 
 
+class _FakeEmbedding:
+    def is_available(self) -> bool:
+        return True
+
+    def embed_text_sync(self, text: str) -> list[float]:
+        return [0.25, 0.5, 0.75]
+
+
+def test_retrieve_evidence_bundle_sync_uses_vector_search(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    searched: list[tuple] = []
+
+    def fake_search(session, user_id, emb, top_k):
+        searched.append((user_id, emb, top_k))
+        return [
+            {
+                "id": "c1",
+                "content": "chunk body",
+                "chunk_index": 0,
+                "distance": 0.1,
+            }
+        ]
+
+    def fake_meta(session, user_id, q, top_k):
+        return {
+            "relevant_facts": [],
+            "timeline_hints": [],
+            "relevant_summaries": [],
+            "relevant_stories": [],
+        }
+
+    monkeypatch.setattr(evidence_mod, "search_chunks_vector_sync", fake_search)
+    monkeypatch.setattr(evidence_mod, "fetch_evidence_metadata_sync", fake_meta)
+
+    out = retrieve_evidence_bundle_sync(
+        session=object(),
+        user_id="u1",
+        query="  hello  ",
+        top_k=7,
+        embedding_provider=_FakeEmbedding(),
+    )
+    assert len(searched) == 1
+    assert searched[0][0] == "u1"
+    assert searched[0][1] == [0.25, 0.5, 0.75]
+    assert searched[0][2] == 7
+    assert out["relevant_chunks"] == [
+        {"id": "c1", "content": "chunk body", "chunk_index": 0},
+    ]
+
+
 def test_empty_evidence_bundle_keys() -> None:
     assert set(EMPTY_EVIDENCE_BUNDLE.keys()) == {
         "relevant_chunks",