Optimize search: parallelize BM25 and embedding, fix word filter

- Run BM25 and embedding searches in parallel with asyncio.gather
- Fix word filter in both search.py and bm25.py: >= 2 chars instead of > 2
  (allows important 2-letter terms like "AI", "ML" to be included)
- Better error handling for parallel search failures

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

src/memory/api/search/bm25.py

@@ -53,7 +53,7 @@ def build_tsquery(query: str) -> str:
     words = [
         w.strip().lower()
         for w in clean_query.split()
-        if w.strip() and len(w.strip()) > 2 and w.strip().lower() not in _STOPWORDS
+        if w.strip() and len(w.strip()) >= 2 and w.strip().lower() not in _STOPWORDS
     ]
     if not words:
         return ""

src/memory/api/search/search.py

@@ -7,7 +7,6 @@ import logging
 import math
 from collections import defaultdict
 from datetime import datetime, timezone
-from sqlalchemy.orm import load_only
 from memory.common import extract, settings
 from memory.common.db.connection import make_session
 from memory.common.db.models import Chunk, SourceItem
@@ -321,22 +320,31 @@ async def _run_searches(
     use_bm25: bool,
 ) -> dict[str, float]:
     """
-    Run embedding and optionally BM25 searches, returning fused scores.
+    Run embedding and optionally BM25 searches in parallel, returning fused scores.
     """
-    # Run embedding search
+    # Build tasks to run in parallel
-    embedding_scores = await search_chunks_embeddings(
+    embedding_task = search_chunks_embeddings(
         search_data, modalities, internal_limit, filters, timeout
     )
 
-    # Run BM25 search if enabled
-    bm25_scores: dict[str, float] = {}
     if use_bm25:
-        try:
+        # Run both searches in parallel
-            bm25_scores = await search_bm25_chunks(
+        results = await asyncio.gather(
-                data, modalities, internal_limit, filters, timeout
+            embedding_task,
+            search_bm25_chunks(data, modalities, internal_limit, filters, timeout),
+            return_exceptions=True,
         )
-        except asyncio.TimeoutError:
+
-            logger.warning("BM25 search timed out, using embedding results only")
+        embedding_scores = results[0] if not isinstance(results[0], Exception) else {}
+        if isinstance(results[0], Exception):
+            logger.warning(f"Embedding search failed: {results[0]}")
+
+        bm25_scores = results[1] if not isinstance(results[1], Exception) else {}
+        if isinstance(results[1], Exception):
+            logger.warning(f"BM25 search failed: {results[1]}")
+    else:
+        embedding_scores = await embedding_task
+        bm25_scores = {}
 
     # Fuse scores from both methods using Reciprocal Rank Fusion
     return fuse_scores_rrf(embedding_scores, bm25_scores)
@@ -367,14 +375,6 @@ def _fetch_chunks(
     with make_session() as db:
         chunks = (
             db.query(Chunk)
-            .options(
-                load_only(
-                    Chunk.id,  # type: ignore
-                    Chunk.source_id,  # type: ignore
-                    Chunk.content,  # type: ignore
-                    Chunk.file_paths,  # type: ignore
-                )
-            )
             .filter(Chunk.id.in_(top_ids))
             .all()
         )