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/app/tasks/memory_compaction_tasks.py

@@ -15,14 +15,33 @@ from app.core.memory_compaction_schedule import (
     finalize_memory_compaction_run,
     read_debounce_deadline_ts,
     release_scheduler_gate,
+    schedule_memory_compaction_run,
     set_incremental_cursor_pair,
 )
 from app.core.redis_lock import acquire_redis_lock, release_redis_lock
 from app.features.memory.compaction_service import run_memory_compaction_sync
+from app.features.memory.repo import list_users_with_recent_chunks_sync
 
 logger = get_logger(__name__)
 
 
+@shared_task
+def memory_compaction_sweep() -> dict[str, Any]:
+    """Beat：为近期有记忆写入的用户调度 compaction（debounce 仍由 schedule 合并）。"""
+    if not settings.memory_compaction_enabled:
+        return {"skipped": True, "reason": "disabled"}
+    hours = int(settings.memory_compaction_sweep_recent_hours)
+    with get_sync_db() as session:
+        user_ids = list_users_with_recent_chunks_sync(session, hours=hours)
+    ctx_base: dict[str, Any] = {"trigger_source": "beat", "sweep_hours": hours}
+    for uid in user_ids:
+        schedule_memory_compaction_run(uid, dict(ctx_base))
+    logger.info(
+        "memory_compaction_sweep hours={} scheduled_users={}", hours, len(user_ids)
+    )
+    return {"scheduled": len(user_ids), "user_ids": user_ids}
+
+
 @shared_task(bind=True, max_retries=12, default_retry_delay=20)
 def memory_compaction_run(
     self, user_id: str, context: dict[str, Any] | None = None