memory evolution phase 1

app/core/memory_middleware.py

@@ -69,17 +69,19 @@ class MemoryMiddleware:
         if fernet is None:
             return {}
 
+        user_dbg = await self._get_user_debug(user_id)
+        user_tier: str = user_dbg.get("tier") or "free"
+
         core = await self._load_core(user_id, fernet)
-        associative = await self._load_associative(user_id, message, fernet)
+        associative = await self._load_associative(user_id, message, fernet, user_tier=user_tier)
         episodic = await self._load_episodic(user_id, fernet, session_id=session_id)
         proactive = await self._load_proactive(user_id, fernet)
 
-        user_dbg = await self._get_user_debug(user_id)
         logger.info(
             "memory: enrich_context trace=%s user=%s tier=%s core=%d associative=%d episodic=%d proactive=%d",
             trace_id or "-",
             user_id,
-            user_dbg.get("tier") or "-",
+            user_tier,
             len(core),
             len(associative),
             len(episodic),
@@ -255,6 +257,50 @@ class MemoryMiddleware:
         logger.info("memory: replace_core user=%s label=%s changed=1", user_id, label)
         return True
 
+    async def store_associative(
+        self,
+        user_id: str,
+        content: str,
+        entity_type: str | None = None,
+        entity_id: str | None = None,
+    ) -> None:
+        """Store associative memory; embed if user tier has real_embeddings."""
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+        from app.core.embeddings import embed_text  # noqa: PLC0415
+
+        fernet = await self._get_fernet(user_id)
+        if fernet is None:
+            return
+
+        encrypted = _encrypt(fernet, content)
+
+        user_dbg = await self._get_user_debug(user_id)
+        user_tier = user_dbg.get("tier") or "free"
+
+        embedding: list[float] | None = None
+        if tier_manager.check_feature(user_tier, "real_embeddings"):
+            embedding = await embed_text(content)
+
+        row = MemoryAssociative(
+            id=str(uuid.uuid4()),
+            user_id=user_id,
+            content_encrypted=encrypted,
+            embedding=embedding,
+            entity_type=entity_type,
+            entity_id=entity_id,
+        )
+        self._db.add(row)
+        try:
+            await self._db.commit()
+            logger.info(
+                "memory: store_associative user=%s embedded=%s",
+                user_id,
+                embedding is not None,
+            )
+        except Exception as exc:
+            logger.error("memory: store_associative failed user=%s: %s", user_id, exc)
+            await self._db.rollback()
+
     async def insert_archival(self, user_id: str, content: str, source: str = "manual") -> None:
         """Insert a long-term archival memory entry."""
         fernet = await self._get_fernet(user_id)
@@ -364,14 +410,49 @@ class MemoryMiddleware:
         return out
 
     async def _load_associative(
-        self, user_id: str, message: str, fernet: Fernet
+        self, user_id: str, message: str, fernet: Fernet, *, user_tier: str = "free"
     ) -> list[str]:
         """Load top-k associative memories.
 
-        Production: uses pgvector cosine similarity on the message embedding.
-        Current implementation: keyword-based fallback (no external embedding call)
-        so tests pass without a live OpenAI key.
+        Pro+: pgvector cosine similarity on the message embedding (real_embeddings feature).
+        Free / embedding failure: keyword-ordered fallback (most recent rows).
         """
+        from app.billing.tier_manager import tier_manager  # noqa: PLC0415
+        from app.core.embeddings import embed_text  # noqa: PLC0415
+
+        if tier_manager.check_feature(user_tier, "real_embeddings"):
+            vec = await embed_text(message)
+            if vec is not None:
+                try:
+                    result = await self._db.execute(
+                        select(MemoryAssociative)
+                        .where(
+                            MemoryAssociative.user_id == user_id,
+                            MemoryAssociative.embedding.isnot(None),
+                        )
+                        .order_by(MemoryAssociative.embedding.cosine_distance(vec))
+                        .limit(_ASSOCIATIVE_TOP_K)
+                    )
+                    rows = result.scalars().all()
+                    out: list[str] = []
+                    for row in rows:
+                        plaintext = _safe_decrypt(fernet, row.content_encrypted)
+                        if plaintext is not None:
+                            out.append(plaintext)
+                    logger.info(
+                        "memory: _load_associative user=%s mode=vector hits=%d",
+                        user_id,
+                        len(out),
+                    )
+                    return out
+                except Exception as exc:
+                    logger.warning(
+                        "memory: vector search failed user=%s, falling back to keyword: %s",
+                        user_id,
+                        exc,
+                    )
+
+        # Keyword fallback: most recent rows
         result = await self._db.execute(
             select(MemoryAssociative)
             .where(MemoryAssociative.user_id == user_id)
@@ -379,7 +460,7 @@ class MemoryMiddleware:
             .limit(_ASSOCIATIVE_TOP_K)
         )
         rows = result.scalars().all()
-        out: list[str] = []
+        out = []
         for row in rows:
             plaintext = _safe_decrypt(fernet, row.content_encrypted)
             if plaintext is not None: