superlocalmemory 3.7.1 → 3.7.3

package/CHANGELOG.md

@@ -5,6 +5,25 @@ All notable changes to SuperLocalMemory V3 will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.7.3] - 2026-07-18 — Scale Engine integrity release
+
+### Fixed
+
+- Added explicit, structurally verified adoption for pre-v3.7 CozoDB/LanceDB projections.
+- Added durable promotion recovery, lifecycle serialization, and a final canonical-source consistency fence.
+- Made Scale Engine status distinguish installed projection paths, lifecycle state, and live daemon backend health.
+- Unified shipped runtime identity metadata at the release version.
+
+## [3.7.2] - 2026-07-16 — Reliability release
+
+### Fixed
+
+- Strengthened daemon-owned local write coordination across Mesh and MCP paths.
+- Kept durable facts queryable when optional enrichment needs a controlled retry.
+- Preserved graph-aware retrieval through the canonical fallback when sqlite-vec is unavailable.
+- Bounded local embedding-worker dependency failures to prevent repeated worker respawns.
+- Hardened Windows RAM reservations and cross-platform installer validation.
+
 ## [3.7.1] - 2026-07-16 — Installer-parity hotfix
 
 ### Fixed

package/README.md

@@ -1,11 +1,14 @@
 <p align="center">
-  <img src="https://superlocalmemory.com/assets/logo-mark.png" alt="SuperLocalMemory" width="200"/>
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="assets/branding/slm-wordmark-dark.svg">
+    <img src="assets/branding/slm-wordmark-light.svg" alt="SuperLocalMemory" width="390">
+  </picture>
 </p>
 
-<h1 align="center">SuperLocalMemory V3.7.1</h1>
+<h1 align="center">SuperLocalMemory V3.7.3</h1>
 <p align="center"><strong>Cache. Compress. Remember. Three surfaces — proxy, MCP tools, or skill. Every setup covered.</strong><br/>
 <em>Local-first agent memory with explicit operating modes, auditable retrieval, and optional Optimize tools.</em></p>
-<p align="center"><code>v3.7.1</code> — <strong>Installer-parity hotfix; release witnesses required before outreach.</strong><br/>
+<p align="center"><code>v3.7.3</code> — <strong>Scale Engine integrity release: explicit legacy adoption, durable promotion recovery, and honest runtime status.</strong><br/>
 Proxy: <code>slm wrap claude</code> &nbsp;·&nbsp; MCP: add <code>slm_compress</code> to your config &nbsp;·&nbsp; Skill: zero-config</p>
 <p align="center"><strong>3 public research preprints</strong> (arXiv + Zenodo archives) · <a href="https://arxiv.org/abs/2603.02240">arXiv:2603.02240</a> · <a href="https://arxiv.org/abs/2603.14588">arXiv:2603.14588</a> · <a href="https://arxiv.org/abs/2604.04514">arXiv:2604.04514</a></p>
 
@@ -108,7 +111,7 @@ health surfaces expose the stages actually completed by the installed runtime.
 | **Retrieval and recall** | Semantic, lexical, temporal, Hopfield and spreading-activation candidate channels; RRF fusion, optional reranking and graph score enhancement | Healthy channels participate; response provenance states the evidence used. |
 | **Brain and learning** | Behavioral patterns, feedback/outcome records, rewards, consolidation, LightGBM-related ranking components, soft prompts, and guarded skill-evolution workflows | Learning is evidence-driven; it does not claim autonomous correctness or guaranteed improvement. |
 | **Knowledge graph and entities** | Canonical entities, aliases, entity profiles, graph edges, scenes, timelines, explorer and graph APIs | Stored/derived graph data is evidence, not an instruction authority. |
-| **Scale Engine** | SQLite + sqlite-vec are canonical. CozoDB graph and LanceDB vector projections are packaged and managed with prepare → verify → promote → rollback | Promotion is explicit and parity-gated; do not advertise an unverified projection as the source of truth. |
+| **Scale Engine** | SQLite + sqlite-vec are canonical. CozoDB graph and LanceDB vector projections are managed with prepare → verify → promote → rollback; a structurally detected pre-v3.7 projection can be explicitly adopted. | Promotion is parity-gated and crash-recoverable. Legacy adoption preserves the prior projection as a rollback backup; if proof fails, routing remains on canonical SQLite and status reports repair required. |
 | **Optimize** | Exact cache, tagged invalidation, safe compression, opt-in aggressive prose compression, CCR originals, proxy/MCP/skill surfaces | Only proxy intercepts a primary provider turn. MCP/skill cache results explicitly routed through SLM. |
 | **Mesh** | Authenticated peer messages, inbox/outbox, locks, offline queue, optional discovery and mesh MCP tools | Mesh is coordination, not automatic replicated memory or conflict resolution. |
 | **Governance and operations** | Provenance, audit/retention/policy surfaces, export/erasure controls, diagnostics, health, backups and daemon lifecycle | These are engineering controls, not a legal certification. |
@@ -227,10 +230,15 @@ optional reranking, and entity-graph score enhancement. The entity graph does
 not create an independent candidate in the current implementation. Core memory
 is SQLite-backed. SQLite and sqlite-vec remain the canonical source of truth.
 The packaged Scale Engine can maintain CozoDB graph and LanceDB vector
-projections, and it remains outside active retrieval paths until
-`slm db scale prepare`, `verify`, and `promote` prove parity against the
-canonical store. This makes the capability available on a fresh installation
-without silently migrating an existing user's data.
+projections, and it remains outside active retrieval paths until a staged
+parity witness proves it matches the canonical store. New installations remain
+on Local Core. During upgrade, `slm db scale status` can identify a positive
+pre-v3.7 layout candidate; the operator confirms it with `slm db scale adopt`.
+SLM then rebuilds from canonical SQLite, verifies it, and promotes it with a
+durable recovery journal while retaining the prior directories as a rollback
+backup. `adopt` reports `restart_required: true`; run `slm restart` before
+checking daemon health. If proof fails, recall remains on SQLite and status
+reports repair required.
 
 Canonical ingestion is a durable state machine: `raw → queryable → enriching →
 complete`, with `failed` retaining raw evidence, error details, attempt count,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "3.7.1",
+  "version": "3.7.3",
   "description": "Local-first agent memory with MCP and an agent-native CLI. Documented clients include Claude Code, Cursor, and Windsurf.",
   "keywords": [
     "ai-memory",

package/plugin/.claude-plugin/plugin.json

@@ -16,5 +16,5 @@
   "mcpServers": "./.mcp.json",
   "name": "superlocalmemory",
   "repository": "https://github.com/qualixar/superlocalmemory",
-  "version": "3.7.1"
+  "version": "3.7.3"
 }

package/plugin/requirements.txt

@@ -1 +1 @@
-superlocalmemory==3.7.1
+superlocalmemory==3.7.3

package/plugin/scripts/slm-launch

@@ -4,12 +4,20 @@
 # Cross-platform counterpart: slm-launch.bat (Windows)
 # Referenced by plugin/.mcp.json as the MCP server command.
 #
-# Resolves the correct venv binary for POSIX:
-#   ${CLAUDE_PLUGIN_DATA}/venv/bin/slm mcp
+# Resolves the correct venv binary for POSIX and joins the namespace daemon
+# before opening the MCP stdio transport. This preserves one writer for all
+# parallel Claude sessions that share CLAUDE_PLUGIN_DATA.
 #
 # On Windows, Claude Code invokes slm-launch.bat instead (same-stem, .bat extension).
 #
 # Environment:
 #   CLAUDE_PLUGIN_DATA — persistent data dir where venv lives
 
-exec "${CLAUDE_PLUGIN_DATA}/venv/bin/slm" mcp
+SLM_BIN="${CLAUDE_PLUGIN_DATA}/venv/bin/slm"
+
+if ! "${SLM_BIN}" serve start >&2; then
+    echo "SLM plugin: unable to start the owned daemon; refusing a direct MCP writer." >&2
+    exit 1
+fi
+
+exec "${SLM_BIN}" mcp

package/plugin/scripts/slm-launch.bat

@@ -4,8 +4,8 @@
 :: Cross-platform counterpart: slm-launch (POSIX bash)
 :: Referenced by plugin/.mcp.json as the MCP server command (Windows picks .bat automatically).
 ::
-:: Resolves the correct venv binary for Windows:
-::   %CLAUDE_PLUGIN_DATA%\venv\Scripts\slm.exe mcp
+:: Resolves the correct venv binary for Windows and joins the namespace daemon
+:: before opening MCP, preserving one writer for parallel Claude sessions.
 ::
 :: On Windows, Python venv places entry points in Scripts\ (not bin\ like POSIX).
 :: This launcher bridges the path difference so ONE .mcp.json command field works
@@ -14,4 +14,10 @@
 :: Environment:
 ::   CLAUDE_PLUGIN_DATA — persistent data dir where venv lives
 
+"%CLAUDE_PLUGIN_DATA%\venv\Scripts\slm.exe" serve start 1>&2
+if errorlevel 1 (
+    echo SLM plugin: unable to start the owned daemon; refusing a direct MCP writer. 1>&2
+    exit /b 1
+)
+
 "%CLAUDE_PLUGIN_DATA%\venv\Scripts\slm.exe" mcp

package/plugin-src/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "3.7.1",
+  "version": "3.7.3",
   "pluginName": "superlocalmemory",
   "displayName": "SuperLocalMemory",
   "repository": "https://github.com/qualixar/superlocalmemory",

package/plugin-src/requirements.txt

@@ -1 +1 @@
-superlocalmemory==3.7.1
+superlocalmemory==3.7.3

package/plugin-src/scripts/slm-launch

@@ -4,12 +4,20 @@
 # Cross-platform counterpart: slm-launch.bat (Windows)
 # Referenced by plugin/.mcp.json as the MCP server command.
 #
-# Resolves the correct venv binary for POSIX:
-#   ${CLAUDE_PLUGIN_DATA}/venv/bin/slm mcp
+# Resolves the correct venv binary for POSIX and joins the namespace daemon
+# before opening the MCP stdio transport. This preserves one writer for all
+# parallel Claude sessions that share CLAUDE_PLUGIN_DATA.
 #
 # On Windows, Claude Code invokes slm-launch.bat instead (same-stem, .bat extension).
 #
 # Environment:
 #   CLAUDE_PLUGIN_DATA — persistent data dir where venv lives
 
-exec "${CLAUDE_PLUGIN_DATA}/venv/bin/slm" mcp
+SLM_BIN="${CLAUDE_PLUGIN_DATA}/venv/bin/slm"
+
+if ! "${SLM_BIN}" serve start >&2; then
+    echo "SLM plugin: unable to start the owned daemon; refusing a direct MCP writer." >&2
+    exit 1
+fi
+
+exec "${SLM_BIN}" mcp

package/plugin-src/scripts/slm-launch.bat

@@ -4,8 +4,8 @@
 :: Cross-platform counterpart: slm-launch (POSIX bash)
 :: Referenced by plugin/.mcp.json as the MCP server command (Windows picks .bat automatically).
 ::
-:: Resolves the correct venv binary for Windows:
-::   %CLAUDE_PLUGIN_DATA%\venv\Scripts\slm.exe mcp
+:: Resolves the correct venv binary for Windows and joins the namespace daemon
+:: before opening MCP, preserving one writer for parallel Claude sessions.
 ::
 :: On Windows, Python venv places entry points in Scripts\ (not bin\ like POSIX).
 :: This launcher bridges the path difference so ONE .mcp.json command field works
@@ -14,4 +14,10 @@
 :: Environment:
 ::   CLAUDE_PLUGIN_DATA — persistent data dir where venv lives
 
+"%CLAUDE_PLUGIN_DATA%\venv\Scripts\slm.exe" serve start 1>&2
+if errorlevel 1 (
+    echo SLM plugin: unable to start the owned daemon; refusing a direct MCP writer. 1>&2
+    exit /b 1
+)
+
 "%CLAUDE_PLUGIN_DATA%\venv\Scripts\slm.exe" mcp

package/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "3.7.1"
+version = "3.7.3"
 description = "Local-first agent memory with auditable hybrid retrieval"
 readme = "README.md"
 license = "AGPL-3.0-or-later"

package/src/superlocalmemory/__init__.py

@@ -32,7 +32,7 @@ if "OMP_NUM_THREADS" not in os.environ:
     os.environ["OMP_NUM_THREADS"] = "2"
 # ---------------------------------------------------------------------------
 
-__version__ = "3.7.1"
+__version__ = "3.7.3"
 
 _REQUIRED_VERSIONS = {
     "sentence_transformers": "5.3.0",

package/src/superlocalmemory/cli/main.py

@@ -270,8 +270,8 @@ def main() -> None:
     )
     db_scale_p.add_argument(
         "scale_action",
-        choices=("status", "prepare", "verify", "promote", "rollback"),
-        help="Lifecycle action. Prepare and verify never mutate active projection paths.",
+        choices=("status", "adopt", "prepare", "verify", "promote", "rollback"),
+        help="Lifecycle action. Adopt confirms a detected legacy projection; prepare and verify never mutate active paths.",
     )
     db_scale_p.add_argument("--stage-id", help="Stage identifier required by verify/promote")
     db_scale_p.add_argument("--backup-id", help="Backup identifier required by rollback")

package/src/superlocalmemory/cli/scale_engine_cmd.py

@@ -9,11 +9,23 @@ def cmd_db_scale(args: Namespace) -> int:
     from superlocalmemory.core.config import SLMConfig
     from superlocalmemory.core.scale_engine import ScaleEngineError, ScaleEngineManager
 
-    manager = ScaleEngineManager(SLMConfig.load())
     action = args.scale_action
+    # Scale Engine projections are currently canonical-default-profile data.
+    # Adoption must not fail merely because the operator is working in another
+    # profile, and it must not rewrite that selected profile.
+    manager = ScaleEngineManager(
+        SLMConfig.load(), profile_id="default" if action == "adopt" else None
+    )
     try:
         if action == "status":
             result = manager.status()
+        elif action == "adopt":
+            result = manager.adopt_legacy_projection()
+            if result is None:
+                raise ScaleEngineError(
+                    "no confirmed legacy projection candidate; see `slm db scale status`"
+                )
+            result = {**result, "restart_required": True}
         elif action == "prepare":
             result = manager.prepare()
         elif action == "verify":

package/src/superlocalmemory/core/backend_orchestrator.py

@@ -97,6 +97,8 @@ class BackendOrchestrator:
         except Exception as exc:
             logger.warning("Initial rebalance failed (non-fatal): %s", exc)
 
+        self._recover_interrupted_scale_promotion()
+
         # Backends may be installed with the product, but installing a wheel
         # is not authorization to mutate an existing data root.  Only a
         # verified, explicit promotion may initialize and migrate projections.
@@ -134,6 +136,19 @@ class BackendOrchestrator:
                      "active" if self._cozo and self._cozo_status() == "active" else "off",
                      "active" if self._lancedb and self._lancedb_status() == "active" else "off")
 
+    def _recover_interrupted_scale_promotion(self) -> None:
+        """Repair an interrupted promotion; never auto-mutate a legacy root."""
+        try:
+            from superlocalmemory.core.scale_engine import ScaleEngineManager
+
+            result = ScaleEngineManager(self._config, profile_id="default").recover_interrupted_promotion()
+            if result:
+                logger.warning("Scale Engine promotion recovery: %s", result)
+        except Exception as exc:
+            # A scale projection is derived data. Startup must keep serving
+            # canonical SQLite even if optional recovery itself is unhealthy.
+            logger.error("Scale Engine recovery requires repair; Local Core remains active: %s", exc)
+
     # ------------------------------------------------------------------
     # Incremental Sync (F-04: called from store_pipeline)
     # ------------------------------------------------------------------

package/src/superlocalmemory/core/embedding_worker.py

@@ -70,7 +70,13 @@ def _load_embedding_model(name: str) -> tuple:
 
     Returns (model, backend_name) or (None, "").
     """
-    from sentence_transformers import SentenceTransformer
+    try:
+        from sentence_transformers import SentenceTransformer
+    except Exception:
+        # Dependency/version errors must stay inside the JSON-lines protocol.
+        # Letting this import escape kills the worker before it can explain the
+        # failure, which the parent previously mislabeled as a long timeout.
+        return None, ""
 
     for backend in _embedding_backend_order():
         try:

package/src/superlocalmemory/core/embeddings.py

@@ -295,6 +295,8 @@ class EmbeddingService:
         never hangs indefinitely on cold model loads or network issues.
         """
         with self._lock:
+            if not self._available:
+                return None
             # Worker recycling: restart after N requests to prevent
             # C++ allocator fragmentation over long-running sessions.
             if self._request_count >= _WORKER_RECYCLE_AFTER and self._worker_proc is not None:
@@ -339,6 +341,12 @@ class EmbeddingService:
                 resp = json.loads(resp_line)
                 if not resp.get("ok"):
                     logger.warning("Worker error: %s", resp.get("error"))
+                    # A well-formed worker error is a terminal local
+                    # dependency/model failure, not a transient pipe race.
+                    # Disable this service and terminate the child so every
+                    # recall does not respawn a heavyweight failing process.
+                    self._available = False
+                    self._kill_worker()
                     return None
                 self._reset_idle_timer()
                 self._request_count += 1

package/src/superlocalmemory/core/engine.py

@@ -24,6 +24,7 @@ from typing import Any
 from superlocalmemory.core.config import CANONICAL_RECALL_LIMIT, SLMConfig
 from superlocalmemory.core.engine_capabilities import Capabilities, CapabilityError
 from superlocalmemory.core.modes import get_capabilities
+from superlocalmemory.learning.outcome_queue import RecallEvent, enqueue_recall
 from superlocalmemory.storage.models import (
     AtomicFact, MemoryRecord, Mode, RecallResponse,
 )
@@ -664,9 +665,6 @@ class MemoryEngine:
         # recall correctness (LLD-02 §4.9, LLD-08 §4.1).
         if session_id:
             try:
-                from superlocalmemory.learning.outcome_queue import (
-                    RecallEvent, enqueue_recall,
-                )
                 fact_ids = tuple(
                     getattr(r.fact, "fact_id", "") or ""
                     for r in getattr(response, "results", [])

package/src/superlocalmemory/core/engine_wiring.py

@@ -388,11 +388,6 @@ def _init_spreading_activation(
     vector_store: Any,
 ) -> Any | None:
     """Create SpreadingActivation for Phase 3 5th retrieval channel."""
-    # V3.3.21: Guard against None vector_store. Without embeddings, SA's
-    # search() crashes with "'NoneType' has no attribute 'search'".
-    if vector_store is None:
-        logger.debug("SpreadingActivation skipped: no vector_store")
-        return None
     try:
         from superlocalmemory.retrieval.spreading_activation import (
             SpreadingActivation,

package/src/superlocalmemory/core/ram_lock.py

@@ -15,12 +15,13 @@ Design notes:
   available before we even try — better to defer than thrash.
 - Lock file lives at ``~/.superlocalmemory/ram_lock.sem`` by default,
   tests may monkeypatch ``RAM_LOCK_PATH``.
-- POSIX only; Windows build is deferred per LLD-00 §7 and MASTER-PLAN H-01.
+- Uses a standard-library advisory lock on POSIX and Windows.  The lock is
+  deliberately file-backed so independent daemon, CLI, and worker processes
+  share the same RAM reservation.
 """
 
 from __future__ import annotations
 
-import fcntl
 import os
 import time
 from contextlib import contextmanager
@@ -29,6 +30,11 @@ from typing import Iterator
 
 import psutil
 
+try:  # POSIX advisory locks (macOS/Linux)
+    import fcntl as _fcntl
+except ImportError:  # Windows uses the stdlib byte-range equivalent below.
+    _fcntl = None
+
 # Public override retained for tests and embedders. ``None`` means resolve the
 # lock dynamically so a long-lived process can switch canonical namespaces
 # without retaining the first root it observed.
@@ -44,6 +50,38 @@ def _ram_lock_path() -> Path:
     return state_path("ram_lock.sem")
 
 
+def _acquire_lock(fd: int) -> None:
+    """Acquire one non-blocking cross-process lock for the semaphore."""
+    if _fcntl is not None:
+        _fcntl.flock(fd, _fcntl.LOCK_EX | _fcntl.LOCK_NB)
+        return
+
+    import msvcrt
+
+    # ``msvcrt.locking`` cannot reserve an empty file.  The marker byte is
+    # purely structural; the human-readable audit marker is written after the
+    # reservation is acquired.
+    if os.fstat(fd).st_size == 0:
+        os.write(fd, b"\0")
+    os.lseek(fd, 0, os.SEEK_SET)
+    try:
+        msvcrt.locking(fd, msvcrt.LK_NBLCK, 1)
+    except OSError as exc:
+        raise BlockingIOError("RAM reservation is already held") from exc
+
+
+def _release_lock(fd: int) -> None:
+    """Release a reservation acquired by :func:`_acquire_lock`."""
+    if _fcntl is not None:
+        _fcntl.flock(fd, _fcntl.LOCK_UN)
+        return
+
+    import msvcrt
+
+    os.lseek(fd, 0, os.SEEK_SET)
+    msvcrt.locking(fd, msvcrt.LK_UNLCK, 1)
+
+
 @contextmanager
 def ram_reservation(
     name: str,
@@ -96,7 +134,7 @@ def ram_reservation(
         deadline = time.time() + timeout_s
         while True:
             try:
-                fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
+                _acquire_lock(fd)
                 break
             except BlockingIOError:
                 if time.time() >= deadline:
@@ -114,7 +152,7 @@ def ram_reservation(
         yield
     finally:
         try:
-            fcntl.flock(fd, fcntl.LOCK_UN)
+            _release_lock(fd)
         finally:
             os.close(fd)