@pentatonic-ai/ai-agent-sdk 0.7.5 → 0.7.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pentatonic-ai/ai-agent-sdk",
-  "version": "0.7.5",
+  "version": "0.7.6",
   "description": "TES SDK — LLM observability and lifecycle tracking via Pentatonic Thing Event System. Track token usage, tool calls, and conversations. Manage things through event-sourced lifecycle stages with AI enrichment and vector search.",
   "type": "module",
   "main": "./dist/index.cjs",

package/packages/memory-engine/compat/server.py

@@ -420,12 +420,19 @@ async def health():
     if failures:
         out["status"] = "degraded" if failures < 3 else "down"
 
-    # Memory count: query L6 doc-store as authoritative.
+    # Memory count: query L6 doc-store as authoritative. L6 /stats
+    # returns vector_chunks (Milvus) and fts_chunks (sqlite content
+    # table). Under healthy operation they're equal — take the max so
+    # the count is honest if one side is mid-rebuild.
     try:
         r = await _client().get(f"{L6_DOC_URL}/stats", timeout=3.0)
         if r.status_code == 200:
             stats = r.json()
-            out["memories"] = stats.get("total_chunks") or stats.get("fts_chunks") or 0
+            out["memories"] = max(
+                int(stats.get("vector_chunks") or 0),
+                int(stats.get("fts_chunks") or 0),
+                int(stats.get("total_chunks") or 0),
+            )
     except Exception:
         out["memories"] = None
     return out

package/packages/memory-engine/docker-compose.test.yml

@@ -29,20 +29,26 @@ services:
       retries: 20
       start_period: 5s
 
+  # Pin the embedding dim explicitly across layers, independent of any
+  # developer-local .env (which may set EMBED_DIM=768 for Ollama-based
+  # local dev). The stub returns 4096; layers must agree.
   l4:
     environment:
       L4_NV_EMBED_URL: http://embed-stub:8041/v1/embeddings
       L4_EMBED_API_KEY: ""
+      L4_EMBED_DIM: "4096"
 
   l5:
     environment:
       L5_NV_EMBED_URL: http://embed-stub:8041/v1/embeddings
       L5_EMBED_API_KEY: ""
+      L5_EMBED_DIM: "4096"
 
   l6:
     environment:
       L6_NV_EMBED_URL: http://embed-stub:8041/v1/embeddings
       L6_EMBED_API_KEY: ""
+      L6_EMBED_DIM: "4096"
 
   l2:
     environment:

package/packages/memory-engine/engine/services/l6/l6-document-store.py

@@ -911,16 +911,34 @@ def serve(port: int = DEFAULT_PORT):
             milvus.insert(collection_name=COLLECTION_NAME, data=rows)
         insert_ms = (_time.time() - t1) * 1000.0
 
-        # Single FTS write (best-effort — search still works without it).
+        # Single FTS write — into the `chunks` content table; the
+        # AFTER INSERT trigger replicates rows into the chunks_fts
+        # virtual table so BM25 search (which JOINs chunks ON rowid)
+        # actually finds them. Earlier versions wrote directly to
+        # chunks_fts, leaving `chunks` empty — which made BM25 return
+        # zero hits AND broke the /stats fts_chunks counter.
         try:
             fts_conn = get_fts_db()
             for r, txt in zip(records, texts):
                 rid = r.get("id") or _hashlib.sha1(txt.encode("utf-8")).hexdigest()[:32]
+                chunk_id = f"l6:{rid}:0"[:63]
                 fts_conn.execute(
-                    "INSERT INTO chunks_fts(text, source_file, arena, heading, entities_json) "
-                    "VALUES (?, ?, ?, ?, ?)",
-                    (txt, (r.get("source_file") or f"{rid}.md"), arena,
-                     (r.get("heading") or ""), "[]"),
+                    "INSERT OR REPLACE INTO chunks "
+                    "(id, text, source_file, arena, doc_type, heading, "
+                    " chunk_index, content_hash, entities_json, indexed_at) "
+                    "VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
+                    (
+                        chunk_id,
+                        txt,
+                        (r.get("source_file") or f"{rid}.md"),
+                        arena,
+                        (r.get("doc_type") or "general"),
+                        (r.get("heading") or ""),
+                        0,
+                        _hashlib.sha1(txt.encode("utf-8")).hexdigest()[:20],
+                        "[]",
+                        now,
+                    ),
                 )
             fts_conn.commit()
             fts_conn.close()