superlocalmemory 3.3.6 → 3.3.7

package/README.md

@@ -3,7 +3,8 @@
 </p>
 
 <h1 align="center">SuperLocalMemory V3.3</h1>
-<p align="center"><strong>The first local-only AI memory to break 74% retrieval on LoCoMo.<br/>No cloud. No APIs. No data leaves your machine.</strong></p>
+<p align="center"><strong>Every other AI forgets. Yours won't.</strong><br/><em>Infinite memory for Claude Code, Cursor, Windsurf & 17+ AI tools.</em></p>
+<p align="center"><code>v3.3.6</code> — Install once. Every session remembers the last. Automatically.</p>
 
 <p align="center">
   <code>+16pp vs Mem0 (zero cloud)</code> &nbsp;·&nbsp; <code>85% Open-Domain (best of any system)</code> &nbsp;·&nbsp; <code>EU AI Act Ready</code>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "3.3.6",
+  "version": "3.3.7",
   "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 = "3.3.6"
+version = "3.3.7"
 description = "Information-geometric agent memory with mathematical guarantees"
 readme = "README.md"
 license = {text = "MIT"}

package/src/superlocalmemory/core/embedding_worker.py

@@ -23,9 +23,10 @@ Part of Qualixar | Author: Varun Pratap Bhardwaj
 from __future__ import annotations
 
 import json
+import os
 import signal
 import sys
-import os
+import threading
 
 # Force CPU BEFORE any torch import
 os.environ["CUDA_VISIBLE_DEVICES"] = ""
@@ -41,8 +42,33 @@ if sys.platform != "win32":
     signal.signal(signal.SIGTERM, lambda *_: sys.exit(0))
 
 
+def _start_parent_watchdog() -> None:
+    """Monitor parent process — self-terminate if parent dies.
+
+    Prevents orphaned workers that consume 500-800 MB each when the parent
+    process crashes, is killed, or exits without cleanup.
+
+    V3.3.7: Added after incident where orphaned workers consumed 33 GB.
+    """
+    parent_pid = os.getppid()
+
+    def _watch() -> None:
+        import time
+        while True:
+            time.sleep(5)
+            try:
+                os.kill(parent_pid, 0)
+            except OSError:
+                os._exit(0)
+
+    t = threading.Thread(target=_watch, daemon=True, name="parent-watchdog")
+    t.start()
+
+
 def _worker_main() -> None:
     """Main loop: read JSON requests from stdin, write responses to stdout."""
+    _start_parent_watchdog()  # V3.3.7: self-terminate if parent dies
+
     import numpy as np
 
     model = None

package/src/superlocalmemory/core/embeddings.py

@@ -15,6 +15,7 @@ Part of Qualixar | Author: Varun Pratap Bhardwaj
 
 from __future__ import annotations
 
+import atexit
 import json
 import logging
 import os
@@ -22,11 +23,15 @@ import subprocess
 import sys
 import threading
 import time
+import weakref
 from pathlib import Path
 from typing import TYPE_CHECKING
 
 import numpy as np
 
+# Track all live embedding services for atexit cleanup
+_live_embedding_services: set[weakref.ref] = set()
+
 if TYPE_CHECKING:
     from numpy.typing import NDArray
 
@@ -69,6 +74,17 @@ class EmbeddingService:
         self._worker_ready = False
         self._request_count: int = 0
 
+        # Register for atexit cleanup (prevent orphaned workers)
+        ref = weakref.ref(self, _live_embedding_services.discard)
+        _live_embedding_services.add(ref)
+
+    def __del__(self) -> None:
+        """Kill worker subprocess when service is garbage-collected."""
+        try:
+            self._kill_worker()
+        except Exception:
+            pass
+
     @property
     def is_available(self) -> bool:
         """Check if embedding service can produce embeddings."""
@@ -338,3 +354,26 @@ class EmbeddingService:
             raise DimensionMismatchError(
                 f"Embedding dimension {actual} != expected {self._config.dimension}"
             )
+
+
+# ---------------------------------------------------------------------------
+# Module-level atexit: kill ALL embedding workers on process exit
+# ---------------------------------------------------------------------------
+
+def _cleanup_all_embedding_services() -> None:
+    """Kill all embedding worker subprocesses on interpreter exit.
+
+    Prevents orphaned 500-800 MB sentence-transformer workers surviving
+    after parent exits (especially during test runs with parallel agents).
+    """
+    for ref in list(_live_embedding_services):
+        svc = ref()
+        if svc is not None:
+            try:
+                svc._kill_worker()
+            except Exception:
+                pass
+    _live_embedding_services.clear()
+
+
+atexit.register(_cleanup_all_embedding_services)

package/src/superlocalmemory/core/recall_worker.py

@@ -20,6 +20,7 @@ import json
 import os
 import signal
 import sys
+import threading
 
 # Force CPU BEFORE any torch import
 os.environ["CUDA_VISIBLE_DEVICES"] = ""
@@ -34,6 +35,29 @@ os.environ["TORCH_DEVICE"] = "cpu"
 if sys.platform != "win32":
     signal.signal(signal.SIGTERM, lambda *_: sys.exit(0))
 
+
+def _start_parent_watchdog() -> None:
+    """Monitor parent process — self-terminate if parent dies.
+
+    Prevents orphaned workers that consume 500+ MB each when the parent
+    process crashes, is killed, or exits without cleanup.
+
+    V3.3.7: Added after incident where orphaned workers consumed 33 GB.
+    """
+    parent_pid = os.getppid()
+
+    def _watch() -> None:
+        import time
+        while True:
+            time.sleep(5)
+            try:
+                os.kill(parent_pid, 0)
+            except OSError:
+                os._exit(0)
+
+    t = threading.Thread(target=_watch, daemon=True, name="parent-watchdog")
+    t.start()
+
 _engine = None
 
 
@@ -209,6 +233,8 @@ def _handle_status() -> dict:
 
 def _worker_main() -> None:
     """Main loop: read JSON requests from stdin, write responses to stdout."""
+    _start_parent_watchdog()  # V3.3.7: self-terminate if parent dies
+
     for line in sys.stdin:
         line = line.strip()
         if not line:

package/src/superlocalmemory/retrieval/reranker.py

@@ -16,6 +16,7 @@ License: MIT
 
 from __future__ import annotations
 
+import atexit
 import json
 import logging
 import os
@@ -23,10 +24,14 @@ import subprocess
 import sys
 import threading
 import time
+import weakref
 from typing import Any
 
 from superlocalmemory.storage.models import AtomicFact
 
+# Track all live reranker instances for atexit cleanup
+_live_rerankers: set[weakref.ref] = set()
+
 logger = logging.getLogger(__name__)
 
 _IDLE_TIMEOUT_SECONDS = 120  # 2 min → kill worker
@@ -64,11 +69,22 @@ class CrossEncoderReranker:
         self._idle_timer: threading.Timer | None = None
         self._request_count: int = 0
 
+        # Register for atexit cleanup (prevent orphaned workers)
+        ref = weakref.ref(self, _live_rerankers.discard)
+        _live_rerankers.add(ref)
+
         # Start background warmup immediately — worker loads model
         # while the rest of init continues. First recall gets instant
         # fallback; second recall uses the warm model.
         self._start_background_warmup()
 
+    def __del__(self) -> None:
+        """Kill worker subprocess when reranker is garbage-collected."""
+        try:
+            self._kill_worker()
+        except Exception:
+            pass
+
     # ------------------------------------------------------------------
     # Background warmup (non-blocking model load)
     # ------------------------------------------------------------------
@@ -330,3 +346,26 @@ class CrossEncoderReranker:
         if resp is None:
             return False
         return resp.get("ok", False)
+
+
+# ---------------------------------------------------------------------------
+# Module-level atexit: kill ALL reranker workers on process exit
+# ---------------------------------------------------------------------------
+
+def _cleanup_all_rerankers() -> None:
+    """Kill all reranker worker subprocesses on interpreter exit.
+
+    Prevents orphaned 1.3 GB ONNX/PyTorch workers surviving after
+    parent exits (especially during test runs with parallel agents).
+    """
+    for ref in list(_live_rerankers):
+        reranker = ref()
+        if reranker is not None:
+            try:
+                reranker._kill_worker()
+            except Exception:
+                pass
+    _live_rerankers.clear()
+
+
+atexit.register(_cleanup_all_rerankers)