Remove unused deps from poetry build, speed test suite to ~16s (#130)

src/brain/embeddings.py

@@ -7,7 +7,6 @@ from __future__ import annotations
 
 import json
 import logging
-import numpy as np
 from typing import List, Union
 
 logger = logging.getLogger(__name__)
@@ -47,7 +46,7 @@ class LocalEmbedder:
             logger.error("sentence-transformers not installed. Run: pip install sentence-transformers")
             raise
     
-    def encode(self, text: Union[str, List[str]]) -> np.ndarray:
+    def encode(self, text: Union[str, List[str]]):
         """Encode text to embedding vector(s).
         
         Args:
@@ -64,20 +63,22 @@ class LocalEmbedder:
     
     def encode_single(self, text: str) -> bytes:
         """Encode single text to bytes for SQLite storage.
-        
+
         Returns:
             Float32 bytes
         """
+        import numpy as np
         embedding = self.encode(text)
         if len(embedding.shape) > 1:
             embedding = embedding[0]
         return embedding.astype(np.float32).tobytes()
     
-    def similarity(self, a: np.ndarray, b: np.ndarray) -> float:
+    def similarity(self, a, b) -> float:
         """Compute cosine similarity between two vectors.
-        
+
         Vectors should already be normalized from encode().
         """
+        import numpy as np
         return float(np.dot(a, b))
 
 

src/brain/memory.py

@@ -106,6 +106,8 @@ class UnifiedMemory:
     def _get_embedder(self):
         """Lazy-load the embedding model."""
         if self._embedder is None:
+            if os.environ.get("TIMMY_SKIP_EMBEDDINGS") == "1":
+                return None
             try:
                 from brain.embeddings import LocalEmbedder
                 self._embedder = LocalEmbedder()

src/timmy/memory/vector_store.py

@@ -29,14 +29,14 @@ def _get_model():
     if _model is not None:
         return _model
     
+    import os
+    # In test mode or low-memory environments, skip embedding model load
+    if os.environ.get("TIMMY_SKIP_EMBEDDINGS") == "1":
+        _has_embeddings = False
+        return None
+
     try:
         from sentence_transformers import SentenceTransformer
-        import os
-        # In test mode or low-memory environments, we might want to skip this
-        if os.environ.get("TIMMY_SKIP_EMBEDDINGS") == "1":
-            _has_embeddings = False
-            return None
-            
         _model = SentenceTransformer('all-MiniLM-L6-v2')
         _has_embeddings = True
         return _model

src/timmy/semantic_memory.py

@@ -37,6 +37,10 @@ def _get_embedding_model():
     """Lazy-load embedding model."""
     global EMBEDDING_MODEL
     if EMBEDDING_MODEL is None:
+        import os
+        if os.environ.get("TIMMY_SKIP_EMBEDDINGS") == "1":
+            EMBEDDING_MODEL = False
+            return EMBEDDING_MODEL
         try:
             from sentence_transformers import SentenceTransformer
             EMBEDDING_MODEL = SentenceTransformer('all-MiniLM-L6-v2')