graph-mind-memory 0.1.3__tar.gz → 0.1.4__tar.gz

{graph_mind_memory-0.1.3/graph_mind_memory.egg-info → graph_mind_memory-0.1.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: graph-mind-memory
-Version: 0.1.3
+Version: 0.1.4
 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

{graph_mind_memory-0.1.3 → graph_mind_memory-0.1.4}/graph_mind_mcp_server.py

@@ -38,8 +38,9 @@ RecordableEventType = Literal[
 # hub over the network whether a newer revision exists, every time it loads. That made the first
 # save of a session take ~95 s and sent requests off the machine for no reason. Once the model is
 # on disk, stay offline.
-if (Path.home() / ".cache" / "huggingface" / "hub"
-        / "models--sentence-transformers--paraphrase-multilingual-MiniLM-L12-v2").is_dir():
+from local_embedder import model_on_disk  # noqa: E402
+
+if model_on_disk():
     os.environ.setdefault("HF_HUB_OFFLINE", "1")
     os.environ.setdefault("TRANSFORMERS_OFFLINE", "1")
 

{graph_mind_memory-0.1.3 → graph_mind_memory-0.1.4/graph_mind_memory.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: graph-mind-memory
-Version: 0.1.3
+Version: 0.1.4
 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

{graph_mind_memory-0.1.3 → graph_mind_memory-0.1.4}/install.py

@@ -159,21 +159,34 @@ def capture_service(start: bool) -> str:
 
 def embedding_model():
     sys.path.insert(0, str(HERE))
-    from local_embedder import LOCAL_MODEL
+    from local_embedder import LOCAL_MODEL, own_model_cache
     try:
         from transformers import AutoModel, AutoTokenizer
-        AutoTokenizer.from_pretrained(LOCAL_MODEL)
-        AutoModel.from_pretrained(LOCAL_MODEL)
     except OSError as error:
+        if "1114" not in str(error):
+            raise
         # WinError 1114 on c10.dll: PyTorch needs the Microsoft Visual C++ runtime, which a fresh
         # Windows often lacks. Everything else is installed; recall uses word search until then.
-        if platform.system() != "Windows":
-            raise
         sys.exit(f"\n{error}\n\nPyTorch could not load. Install the Microsoft Visual C++ "
                  "Redistributable (x64), restart, and run this again:\n"
                  "    https://aka.ms/vs/17/release/vc_redist.x64.exe\n"
                  "The apps are already connected; until then recall works by word search only.")
 
+    def download(cache_dir):
+        AutoTokenizer.from_pretrained(LOCAL_MODEL, cache_dir=cache_dir)
+        AutoModel.from_pretrained(LOCAL_MODEL, cache_dir=cache_dir)
+    try:
+        download(None)
+    except OSError as error:
+        # ~/.cache refused (access denied): keep the model in Graph-MIND's own folder instead,
+        # where the server looks too (local_embedder.model_cache)
+        if "PermissionError" not in str(error) and not isinstance(error.__cause__, PermissionError):
+            raise
+        own = own_model_cache()
+        own.mkdir(parents=True, exist_ok=True)
+        step(f"the usual model folder is not writable; using {own}")
+        download(str(own))
+
 
 def main(argv=None):
     parser = argparse.ArgumentParser(description="Install Graph-MIND on this PC.")

{graph_mind_memory-0.1.3 → graph_mind_memory-0.1.4}/local_embedder.py

@@ -8,6 +8,7 @@ from __future__ import annotations
 from pathlib import Path
 import hashlib
 import math
+import os
 import threading
 
 from associative_memory import _terms
@@ -18,6 +19,24 @@ ACTIVATE = dict(seed_limit=12, hops=2, fanout=8, max_nodes=24, limit=8)  # froze
 SEED_LIMIT = ACTIVATE["seed_limit"]
 
 
+def own_model_cache() -> Path:
+    from development_paths import graph_mind_home
+    return graph_mind_home() / "models"
+
+
+def model_cache():
+    """None means Hugging Face's usual cache. Graph-MIND's own folder once the installer had to
+    use it: on one fresh Windows profile ~/.cache could not be written (access denied)."""
+    own = own_model_cache()
+    return str(own) if own.is_dir() else None
+
+
+def model_on_disk() -> bool:
+    name = "models--" + LOCAL_MODEL.replace("/", "--")
+    usual = Path(os.environ.get("HF_HOME") or Path.home() / ".cache" / "huggingface") / "hub"
+    return (usual / name).is_dir() or (own_model_cache() / name).is_dir()
+
+
 class LocalEmbedder:
     """On-device multilingual embeddings via raw transformers (mean pooling, L2-normalized).
     No text leaves the machine — required by the user-owned local-brain promise."""
@@ -42,8 +61,10 @@ class LocalEmbedder:
         if LocalEmbedder._model is None:
             import torch
             from transformers import AutoModel, AutoTokenizer
-            LocalEmbedder._tok = AutoTokenizer.from_pretrained(self.model_name)
-            LocalEmbedder._model = AutoModel.from_pretrained(self.model_name).eval()
+            cache = model_cache()
+            LocalEmbedder._tok = AutoTokenizer.from_pretrained(self.model_name, cache_dir=cache)
+            LocalEmbedder._model = AutoModel.from_pretrained(self.model_name,
+                                                             cache_dir=cache).eval()
             LocalEmbedder._torch = torch
 
     def _key(self, text):

{graph_mind_memory-0.1.3 → graph_mind_memory-0.1.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "graph-mind-memory"
-version = "0.1.3"
+version = "0.1.4"
 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"