graph-mind-memory 0.3.3__tar.gz → 0.3.4__tar.gz

{graph_mind_memory-0.3.3/graph_mind_memory.egg-info → graph_mind_memory-0.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: graph-mind-memory
-Version: 0.3.3
+Version: 0.3.4
 Summary: Local-first memory for AI coding assistants: verbatim, on your own PC, no LLM calls at write time.
 Author-email: Yohan Ko <goyohan0611@gmail.com>
 License-Expression: AGPL-3.0-only
@@ -258,8 +258,9 @@ python -m unittest discover -p "test_*.py"
 
 ## Known limits
 
-- **The first question after an AI app starts waits for the embedding model to load** (a few
-  seconds; 30-50 s on a slow or synced disk). Later questions take well under a second.
+- **For about the first half minute after an AI app starts, answers use word search only** while the
+  embedding model loads in the background (the first answer takes ~2.5 s instead of waiting out a
+  30-50 s load on a slow or synced disk). After that, questions take well under a second.
 - **Daily use on Windows only so far.** macOS and Linux pass the tests in CI but have not seen real
   use.
 - **Capture follows each app's transcript format,** which is not a public interface. An app update

{graph_mind_memory-0.3.3 → graph_mind_memory-0.3.4}/README.md

@@ -232,8 +232,9 @@ python -m unittest discover -p "test_*.py"
 
 ## Known limits
 
-- **The first question after an AI app starts waits for the embedding model to load** (a few
-  seconds; 30-50 s on a slow or synced disk). Later questions take well under a second.
+- **For about the first half minute after an AI app starts, answers use word search only** while the
+  embedding model loads in the background (the first answer takes ~2.5 s instead of waiting out a
+  30-50 s load on a slow or synced disk). After that, questions take well under a second.
 - **Daily use on Windows only so far.** macOS and Linux pass the tests in CI but have not seen real
   use.
 - **Capture follows each app's transcript format,** which is not a public interface. An app update

{graph_mind_memory-0.3.3 → graph_mind_memory-0.3.4}/graph_mind_mcp_server.py

@@ -218,7 +218,15 @@ def semantic_turns(database, query, *, scopes=None, since=None, until=None, limi
     # new pieces are embedded inline; a backlog is filled in the background while this question
     # is answered from what is already embedded (word search covers the rest meanwhile).
     missing = [p for p in pieces if embedder._key(p.strip()[:2000]) not in embedder.cache]
-    if len(missing) > INLINE_EMBED_LIMIT:
+    # Nor wait for the model itself: an app's first question used to sit through the torch import
+    # (38.8 s measured). If it is not loaded within WAIT_FOR_MODEL seconds, this question is
+    # answered from word search plus what is embedded already, and loading carries on.
+    asked = embedder._key(query.strip()[:2000]) in embedder.cache
+    wait = getattr(embedder, "WAIT_FOR_MODEL", None)
+    loading = (wait is not None and (missing or not asked) and not embedder.ready(wait))
+    if loading and not asked:
+        return []
+    if len(missing) > INLINE_EMBED_LIMIT or (loading and missing):
         embed_in_background(embedder, missing)
         keep = [i for i, p in enumerate(pieces)
                 if embedder._key(p.strip()[:2000]) in embedder.cache]
@@ -741,6 +749,8 @@ def preload(database):
 
 
 def main():
+    from local_embedder import LocalEmbedder
+    LocalEmbedder.WAIT_FOR_MODEL = 0.5      # an app's question never waits out a model load
     preload(configured_db_path())
     server.run(transport="stdio")
 

{graph_mind_memory-0.3.3 → graph_mind_memory-0.3.4/graph_mind_memory.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: graph-mind-memory
-Version: 0.3.3
+Version: 0.3.4
 Summary: Local-first memory for AI coding assistants: verbatim, on your own PC, no LLM calls at write time.
 Author-email: Yohan Ko <goyohan0611@gmail.com>
 License-Expression: AGPL-3.0-only
@@ -258,8 +258,9 @@ python -m unittest discover -p "test_*.py"
 
 ## Known limits
 
-- **The first question after an AI app starts waits for the embedding model to load** (a few
-  seconds; 30-50 s on a slow or synced disk). Later questions take well under a second.
+- **For about the first half minute after an AI app starts, answers use word search only** while the
+  embedding model loads in the background (the first answer takes ~2.5 s instead of waiting out a
+  30-50 s load on a slow or synced disk). After that, questions take well under a second.
 - **Daily use on Windows only so far.** macOS and Linux pass the tests in CI but have not seen real
   use.
 - **Capture follows each app's transcript format,** which is not a public interface. An app update

{graph_mind_memory-0.3.3 → graph_mind_memory-0.3.4}/local_embedder.py

@@ -50,6 +50,12 @@ class LocalEmbedder:
     _lock = threading.RLock()  # model load + cache writes are not thread-safe
 
     FLUSH_EVERY = 5000   # vectors between writes: bounds lost work without rewriting per call
+    # Seconds a question may wait for the model to load before it is answered by word search
+    # alone. None waits (benchmarks, tests, the capture service); the MCP server sets 0.5: loading
+    # torch took 38.8 s for an app's first question on a OneDrive-synced Windows PC; now 2.4 s.
+    WAIT_FOR_MODEL = None
+    _loaded = threading.Event()
+    _loading = threading.Lock()
 
     def __init__(self, cache_path, model=LOCAL_MODEL):
         # float32 rows in a sibling directory: the old JSON cache turned 1.1GB on disk into 3-4GB
@@ -68,6 +74,24 @@ class LocalEmbedder:
             LocalEmbedder._tok = AutoTokenizer.from_pretrained(self.model_name)
             LocalEmbedder._model = AutoModel.from_pretrained(self.model_name).eval()
             LocalEmbedder._torch = torch
+            LocalEmbedder._loaded.set()
+
+    def ready(self, timeout) -> bool:
+        """Whether the model is loaded, waiting at most `timeout` seconds. The first call starts
+        loading it in the background, so a later question finds it ready."""
+        if LocalEmbedder._model is not None:
+            return True
+        if LocalEmbedder._loading.acquire(blocking=False):
+            def load():
+                try:
+                    with LocalEmbedder._lock:
+                        self._load()
+                except Exception:
+                    pass                     # no model on disk: word search keeps answering
+                finally:
+                    LocalEmbedder._loading.release()
+            threading.Thread(target=load, name="graph-mind-model", daemon=True).start()
+        return LocalEmbedder._loaded.wait(timeout)
 
     def _key(self, text):
         return hashlib.sha256((self.model_name + "\0" + text).encode("utf-8")).hexdigest()

{graph_mind_memory-0.3.3 → graph_mind_memory-0.3.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "graph-mind-memory"
-version = "0.3.3"
+version = "0.3.4"
 description = "Local-first memory for AI coding assistants: verbatim, on your own PC, no LLM calls at write time."
 readme = "README.md"
 requires-python = ">=3.10"