superlocalmemory 3.0.17 → 3.0.19

package/bin/slm-npm

@@ -81,6 +81,14 @@ const result = spawnSync(pythonParts[0], [
     env: {
         ...process.env,
         PYTHONPATH: SRC_DIR + (process.env.PYTHONPATH ? path.delimiter + process.env.PYTHONPATH : ''),
+        // Prevent PyTorch Metal/MPS GPU memory reservation on Apple Silicon.
+        // Without these, macOS Activity Monitor shows 3-6 GB for a 40 MB process.
+        PYTORCH_MPS_HIGH_WATERMARK_RATIO: '0.0',
+        PYTORCH_MPS_MEM_LIMIT: '0',
+        PYTORCH_ENABLE_MPS_FALLBACK: '1',
+        TOKENIZERS_PARALLELISM: 'false',
+        TORCH_DEVICE: 'cpu',
+        CUDA_VISIBLE_DEVICES: '',
     },
 });
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "3.0.17",
+  "version": "3.0.19",
   "description": "Information-geometric agent memory with mathematical guarantees. 4-channel retrieval, Fisher-Rao similarity, zero-LLM mode, EU AI Act compliant. Works with Claude, Cursor, Windsurf, and 17+ AI tools.",
   "keywords": [
     "ai-memory",

package/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "3.0.17"
+version = "3.0.19"
 description = "Information-geometric agent memory with mathematical guarantees"
 readme = "README.md"
 license = {text = "MIT"}

package/src/superlocalmemory/cli/commands.py

@@ -27,6 +27,8 @@ def dispatch(args: Namespace) -> None:
         "remember": cmd_remember,
         "recall": cmd_recall,
         "forget": cmd_forget,
+        "delete": cmd_delete,
+        "update": cmd_update,
         "status": cmd_status,
         "health": cmd_health,
         "trace": cmd_trace,
@@ -199,6 +201,72 @@ def cmd_forget(args: Namespace) -> None:
         print("Cancelled.")
 
 
+def cmd_delete(args: Namespace) -> None:
+    """Delete a specific memory by exact fact ID."""
+    from superlocalmemory.core.config import SLMConfig
+    from superlocalmemory.core.engine import MemoryEngine
+
+    config = SLMConfig.load()
+    engine = MemoryEngine(config)
+    engine.initialize()
+
+    fact_id = args.fact_id.strip()
+    # Look up the memory first so user can confirm
+    rows = engine._db.execute(
+        "SELECT content FROM atomic_facts WHERE fact_id = ? AND profile_id = ?",
+        (fact_id, engine.profile_id),
+    )
+    if not rows:
+        print(f"Memory not found: {fact_id}")
+        return
+
+    content_preview = dict(rows[0]).get("content", "")[:120]
+    print(f"Memory: {content_preview}")
+
+    if not getattr(args, "yes", False):
+        confirm = input("Delete this memory? [y/N] ").strip().lower()
+        if confirm not in ("y", "yes"):
+            print("Cancelled.")
+            return
+
+    engine._db.delete_fact(fact_id)
+    print(f"Deleted: {fact_id}")
+
+
+def cmd_update(args: Namespace) -> None:
+    """Update the content of a specific memory by exact fact ID."""
+    from superlocalmemory.core.config import SLMConfig
+    from superlocalmemory.core.engine import MemoryEngine
+
+    config = SLMConfig.load()
+    engine = MemoryEngine(config)
+    engine.initialize()
+
+    fact_id = args.fact_id.strip()
+    new_content = args.content.strip()
+    if not new_content:
+        print("Error: content cannot be empty")
+        return
+
+    rows = engine._db.execute(
+        "SELECT content FROM atomic_facts WHERE fact_id = ? AND profile_id = ?",
+        (fact_id, engine.profile_id),
+    )
+    if not rows:
+        print(f"Memory not found: {fact_id}")
+        return
+
+    old_content = dict(rows[0]).get("content", "")
+    print(f"Old: {old_content[:100]}")
+    print(f"New: {new_content[:100]}")
+
+    engine._db.execute(
+        "UPDATE atomic_facts SET content = ? WHERE fact_id = ?",
+        (new_content, fact_id),
+    )
+    print(f"Updated: {fact_id}")
+
+
 def cmd_status(_args: Namespace) -> None:
     """Show system status."""
     from superlocalmemory.core.config import SLMConfig

package/src/superlocalmemory/cli/main.py

@@ -11,6 +11,16 @@ Part of Qualixar | Author: Varun Pratap Bhardwaj
 
 from __future__ import annotations
 
+# CRITICAL: Set BEFORE any torch/transformers import to prevent Metal/MPS
+# GPU memory reservation on Apple Silicon. Without this, macOS Activity
+# Monitor shows 3-6 GB for what is actually a 40 MB process.
+import os as _os
+_os.environ.setdefault('PYTORCH_MPS_HIGH_WATERMARK_RATIO', '0.0')
+_os.environ.setdefault('PYTORCH_MPS_MEM_LIMIT', '0')
+_os.environ.setdefault('PYTORCH_ENABLE_MPS_FALLBACK', '1')
+_os.environ.setdefault('TOKENIZERS_PARALLELISM', 'false')
+_os.environ.setdefault('TORCH_DEVICE', 'cpu')
+
 import argparse
 import sys
 
@@ -109,10 +119,18 @@ def main() -> None:
     recall_p.add_argument("query", help="Search query")
     recall_p.add_argument("--limit", type=int, default=10, help="Max results (default 10)")
 
-    forget_p = sub.add_parser("forget", help="Delete memories matching a query")
+    forget_p = sub.add_parser("forget", help="Delete memories matching a query (fuzzy)")
     forget_p.add_argument("query", help="Query to match for deletion")
 
-    list_p = sub.add_parser("list", help="List recent memories chronologically")
+    delete_p = sub.add_parser("delete", help="Delete a specific memory by ID (precise)")
+    delete_p.add_argument("fact_id", help="Exact fact ID to delete")
+    delete_p.add_argument("--yes", "-y", action="store_true", help="Skip confirmation")
+
+    update_p = sub.add_parser("update", help="Edit the content of a specific memory by ID")
+    update_p.add_argument("fact_id", help="Exact fact ID to update")
+    update_p.add_argument("content", help="New content for the memory")
+
+    list_p = sub.add_parser("list", help="List recent memories chronologically (shows IDs for delete/update)")
     list_p.add_argument(
         "--limit", "-n", type=int, default=20, help="Number of entries (default 20)",
     )

package/src/superlocalmemory/core/embedding_worker.py

@@ -0,0 +1,120 @@
+# Copyright (c) 2026 Varun Pratap Bhardwaj / Qualixar
+# Licensed under the MIT License - see LICENSE file
+# Part of SuperLocalMemory V3 | https://qualixar.com | https://varunpratap.com
+
+"""Subprocess embedding worker — isolates PyTorch memory from main process.
+
+The main process (dashboard/MCP) stays at ~60 MB. All PyTorch/model memory
+lives in this worker subprocess, which auto-kills after idle timeout.
+
+Protocol (JSON over stdin/stdout):
+  Request:  {"cmd": "embed",  "texts": ["hello"]}
+  Response: {"ok": true, "vectors": [[0.1, ...]], "dim": 768}
+
+  Request:  {"cmd": "ping"}
+  Response: {"ok": true}
+
+  Request:  {"cmd": "quit"}
+  (worker exits)
+
+Part of Qualixar | Author: Varun Pratap Bhardwaj
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+import os
+
+# Force CPU BEFORE any torch import
+os.environ["CUDA_VISIBLE_DEVICES"] = ""
+os.environ["PYTORCH_MPS_HIGH_WATERMARK_RATIO"] = "0.0"
+os.environ["PYTORCH_MPS_MEM_LIMIT"] = "0"
+os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+os.environ["TORCH_DEVICE"] = "cpu"
+
+
+def _worker_main() -> None:
+    """Main loop: read JSON requests from stdin, write responses to stdout."""
+    import numpy as np
+
+    model = None
+    model_name = None
+    dim = 0
+
+    for line in sys.stdin:
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            req = json.loads(line)
+        except json.JSONDecodeError:
+            _respond({"ok": False, "error": "Invalid JSON"})
+            continue
+
+        cmd = req.get("cmd", "")
+
+        if cmd == "quit":
+            break
+
+        if cmd == "ping":
+            _respond({"ok": True})
+            continue
+
+        if cmd == "load":
+            name = req.get("model_name", "nomic-ai/nomic-embed-text-v1.5")
+            expected_dim = req.get("dimension", 768)
+            try:
+                from sentence_transformers import SentenceTransformer
+                model = SentenceTransformer(name, trust_remote_code=True, device="cpu")
+                dim = model.get_sentence_embedding_dimension()
+                if dim != expected_dim:
+                    _respond({"ok": False, "error": f"Dimension mismatch: {dim} != {expected_dim}"})
+                    model = None
+                    continue
+                model_name = name
+                _respond({"ok": True, "dim": dim, "model": name})
+            except Exception as exc:
+                _respond({"ok": False, "error": str(exc)})
+            continue
+
+        if cmd == "embed":
+            texts = req.get("texts", [])
+            if not texts:
+                _respond({"ok": False, "error": "No texts provided"})
+                continue
+            if model is None:
+                # Auto-load if not yet loaded
+                name = req.get("model_name", "nomic-ai/nomic-embed-text-v1.5")
+                expected_dim = req.get("dimension", 768)
+                try:
+                    from sentence_transformers import SentenceTransformer
+                    model = SentenceTransformer(name, trust_remote_code=True, device="cpu")
+                    dim = model.get_sentence_embedding_dimension()
+                    model_name = name
+                except Exception as exc:
+                    _respond({"ok": False, "error": f"Model load failed: {exc}"})
+                    continue
+            try:
+                vecs = model.encode(texts, normalize_embeddings=True)
+                if isinstance(vecs, np.ndarray) and vecs.ndim == 2:
+                    result = [vecs[i].tolist() for i in range(vecs.shape[0])]
+                else:
+                    result = [np.asarray(v, dtype=np.float32).tolist() for v in vecs]
+                _respond({"ok": True, "vectors": result, "dim": dim})
+            except Exception as exc:
+                _respond({"ok": False, "error": str(exc)})
+            continue
+
+        _respond({"ok": False, "error": f"Unknown command: {cmd}"})
+
+
+def _respond(data: dict) -> None:
+    """Write JSON response to stdout, flush immediately."""
+    sys.stdout.write(json.dumps(data) + "\n")
+    sys.stdout.flush()
+
+
+if __name__ == "__main__":
+    _worker_main()