superlocalmemory 3.7.5 → 3.7.7

package/CHANGELOG.md

@@ -5,6 +5,33 @@ 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).
 
+## [3.7.7] - 2026-07-20 — Profile isolation and runtime integrity
+
+### Fixed
+
+- Fixes daemon-aware profile switching and profile-isolated CLI reads/writes. Profile transitions now drain admitted operations, atomically rebind the resident engine, persist only after a successful transition, and return a generation-stamped acknowledgement to CLI, dashboard/API, and MCP callers without restarting the daemon.
+- Dashboard mode, provider, embedding, and memory-visibility changes now take effect through the same daemon-owned runtime transition boundary. Existing custom embedding settings survive setup-mode changes, and Optimize cache vectors follow the configured embedding dimension instead of assuming 768 dimensions.
+- Migration reconciliation now accepts only explicitly allowlisted historical hashes and verifies the complete required schema before updating migration metadata; unknown or structurally incomplete drift fails closed.
+- Restored cross-platform UI test discovery and added dependency and high-severity static security gates to CI.
+- Upgraded the audited web, MCP, cryptography, and Transformers dependency stack to patched releases. Three narrowly scoped NLTK, setuptools, and PyTorch advisories are tracked as exact, dated exceptions; PyTorch remains on the proven 2.11 runtime because the combined native/ML upgrade produced a full-suite process crash and could not be attributed safely to one package.
+
+### Added
+
+- Dashboard controls and API/CLI configuration support for the default write scope and explicit shared/global recall opt-ins. Personal-only recall remains the default.
+
+## [3.7.6] - 2026-07-19 — Auth, upgrade, and embedding-dimension fixes
+
+### Fixed
+
+- Writes authenticated by the daemon capability or the dashboard install token are no longer rejected once an `api_key` file is configured. The write path ran a redundant second gate that only understood `X-SLM-API-Key`, so capability-authenticated MCP `remember` write-throughs and install-token dashboard writes / config tests returned 401 "Invalid or missing API key" whenever opt-in API-key auth was enabled. The mutation-actor gate — which already accepts the daemon capability, the install token, a matching API key, or an uncredentialed loopback caller — is now the single authoritative write boundary. (#71, #73, #74)
+- Upgrading an install across a benign migration DDL change no longer leaves the daemon permanently `not_ready`. On a `ddl_sha256` mismatch for a migration already marked complete, the runner now consults the migration's `verify()`; when the schema end-state is present it reconciles the log to the current hash instead of failing readiness. Real drift with an absent schema is still surfaced as a failure. (#70)
+- The LanceDB vector backend now follows the configured embedding dimension instead of a hardcoded 768, so custom OpenAI-compatible endpoints (for example 1024-d Qwen3-Embedding) no longer hit a vector-dimension mismatch on initialization. An existing store's on-disk width is always honored, keeping already-materialized data readable after a configuration change. (#72)
+
+### Improved
+
+- `slm setup` skips the local 768-d embedding-model download when a remote/OpenAI-compatible embedding endpoint is configured, instead of forcing an unnecessary model fetch. (#72)
+- Aligned retrieval documentation and docstrings with the shipped architecture: five parallel candidate producers (semantic, BM25, temporal, spreading-activation, Hopfield) feed single-pass RRF fusion, followed by optional cross-encoder rerank and an entity-graph post-fusion score enhancement. The entity graph is not a sixth parallel candidate producer.
+
 ## [3.7.5] - 2026-07-18 — Complete Scale Engine projection parity
 
 ### Improved

package/README.md

@@ -5,10 +5,10 @@
   </picture>
 </p>
 
-<h1 align="center">SuperLocalMemory V3.7.5</h1>
+<h1 align="center">SuperLocalMemory V3.7.7</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>v3.7.5</code> — <strong>Scale Engine graph and vector projection parity, durable promotion recovery, and honest runtime status.</strong><br/>
+<p align="center"><code>v3.7.7</code> — <strong>Daemon-aware profile isolation, atomic runtime configuration, and verified migration integrity.</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>
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "3.7.5",
+  "version": "3.7.7",
   "description": "Local-first agent memory with MCP and an agent-native CLI. Documented clients include Claude Code, Cursor, and Windsurf.",
   "keywords": [
     "ai-memory",
@@ -51,7 +51,7 @@
     "prepack": "node scripts/build-plugin.mjs && node scripts/prepack.js",
     "postinstall": "node scripts/postinstall.js",
     "preuninstall": "node scripts/preuninstall.js",
-    "test": "node --test 'tests/ui/*.mjs'"
+    "test": "node scripts/run-ui-tests.mjs"
   },
   "engines": {
     "node": ">=18.0.0",

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

@@ -16,5 +16,5 @@
   "mcpServers": "./.mcp.json",
   "name": "superlocalmemory",
   "repository": "https://github.com/qualixar/superlocalmemory",
-  "version": "3.7.5"
+  "version": "3.7.7"
 }

package/plugin/requirements.txt

@@ -1 +1 @@
-superlocalmemory==3.7.5
+superlocalmemory==3.7.7

package/plugin/skills/slm-recall/SKILL.md

@@ -142,9 +142,10 @@ once you have the `fact_id` for full content.
 
 ## How multi-channel retrieval works
 
-`recall` runs four channels in parallel — semantic vector similarity, lexical
-BM25, temporal recency, and structural/graph — then fuses them with Reciprocal
-Rank Fusion (RRF) and applies a reranker. The `channel_weights` field in the
+`recall` runs five candidate producers in parallel — semantic vector similarity,
+lexical BM25, temporal recency, spreading activation, and Hopfield — then fuses
+them with Reciprocal Rank Fusion (RRF), applies a reranker, and layers an
+optional entity-graph post-fusion score enhancement. The `channel_weights` field in the
 response shows how each channel contributed for that query. Weights adapt over
 time based on engagement signals attributed via `session_id`.
 

package/plugin-src/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "3.7.5",
+  "version": "3.7.7",
   "pluginName": "superlocalmemory",
   "displayName": "SuperLocalMemory",
   "repository": "https://github.com/qualixar/superlocalmemory",

package/plugin-src/requirements.txt

@@ -1 +1 @@
-superlocalmemory==3.7.5
+superlocalmemory==3.7.7

package/plugin-src/skills/slm-recall/SKILL.md

@@ -142,9 +142,10 @@ once you have the `fact_id` for full content.
 
 ## How multi-channel retrieval works
 
-`recall` runs four channels in parallel — semantic vector similarity, lexical
-BM25, temporal recency, and structural/graph — then fuses them with Reciprocal
-Rank Fusion (RRF) and applies a reranker. The `channel_weights` field in the
+`recall` runs five candidate producers in parallel — semantic vector similarity,
+lexical BM25, temporal recency, spreading activation, and Hopfield — then fuses
+them with Reciprocal Rank Fusion (RRF), applies a reranker, and layers an
+optional entity-graph post-fusion score enhancement. The `channel_weights` field in the
 response shows how each channel contributed for that query. Weights adapt over
 time based on engagement signals attributed via `session_id`.
 

package/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "3.7.5"
+version = "3.7.7"
 description = "Local-first agent memory with auditable hybrid retrieval"
 readme = "README.md"
 license = "AGPL-3.0-or-later"
@@ -37,12 +37,12 @@ dependencies = [
     "numpy==2.4.4",
     "scipy==1.17.1",
     "networkx==3.6.1",
-    "mcp==1.27.1",
+    "mcp==1.28.1",
     "python-dateutil==2.9.0.post0",
     "rank-bm25==0.2.2",
     "vadersentiment==3.3.2",
     "einops==0.8.2",
-    "fastapi[all]==0.136.1",
+    "fastapi[all]==0.139.2",
     "uvicorn==0.46.0",
     "websockets==16.0",
     "zeroconf>=0.140",
@@ -59,7 +59,7 @@ dependencies = [
     "psutil==7.2.2",
     "structlog==25.5.0",
     "portalocker==3.2.0",
-    "cryptography==45.0.2",
+    "cryptography==48.0.1",
     # Semantic search + cross-encoder reranker. Do NOT use
     # sentence-transformers[onnx] — its extras pull optimum which
     # overrides the sentence-transformers pin via transitive deps.
@@ -67,8 +67,8 @@ dependencies = [
     "sentence-transformers==5.3.0",
     "optimum==2.1.0",
     "onnxruntime==1.24.4",
-    "transformers==4.57.6",
-    "huggingface_hub==0.36.2",
+    "transformers==5.5.4",
+    "huggingface_hub==1.5.0",
     "torch==2.11.0",
     "scikit-learn==1.8.0",
     # Vector KNN extension for the semantic channel.

package/src/superlocalmemory/__init__.py

@@ -32,7 +32,7 @@ if "OMP_NUM_THREADS" not in os.environ:
     os.environ["OMP_NUM_THREADS"] = "2"
 # ---------------------------------------------------------------------------
 
-__version__ = "3.7.5"
+__version__ = "3.7.7"
 
 _REQUIRED_VERSIONS = {
     "sentence_transformers": "5.3.0",

package/src/superlocalmemory/cli/commands.py

@@ -612,6 +612,8 @@ def cmd_config(args: Namespace) -> None:
             "evolution.enabled", "evolution.backend", "evolution.max_evolutions_per_cycle",
             "mesh_enabled", "daemon_idle_timeout", "entity_compilation_enabled",
             "graph_backend", "vector_backend", "scale_engine_state",
+            "scope.default_scope", "scope.recall_include_global",
+            "scope.recall_include_shared",
         }
         if key not in _ALLOWED_CONFIG_KEYS:
             if use_json:
@@ -650,6 +652,60 @@ def cmd_config(args: Namespace) -> None:
                 except ValueError:
                     parsed_value = value
 
+        if key == "scope.default_scope" and parsed_value not in {
+            "personal", "shared", "global",
+        }:
+            message = "scope.default_scope must be personal, shared, or global"
+            if use_json:
+                from superlocalmemory.cli.json_output import json_print
+                json_print("config", error={
+                    "code": "INVALID_VALUE", "message": message,
+                })
+            else:
+                print(f"Error: {message}")
+            sys.exit(1)
+        if key in {
+            "scope.recall_include_global", "scope.recall_include_shared",
+        } and not isinstance(parsed_value, bool):
+            message = f"{key} must be true or false"
+            if use_json:
+                from superlocalmemory.cli.json_output import json_print
+                json_print("config", error={
+                    "code": "INVALID_VALUE", "message": message,
+                })
+            else:
+                print(f"Error: {message}")
+            sys.exit(1)
+
+        if key.startswith("scope."):
+            from superlocalmemory.cli.daemon import daemon_request, is_daemon_running
+
+            if is_daemon_running():
+                field = key.split(".", 1)[1]
+                result = daemon_request(
+                    "PUT", "/api/v3/scope/config", {field: parsed_value},
+                )
+                if not isinstance(result, dict) or result.get("success") is not True:
+                    message = "resident daemon rejected the scope configuration"
+                    if use_json:
+                        from superlocalmemory.cli.json_output import json_print
+                        json_print("config", error={
+                            "code": "CONFIG_APPLY_FAILED", "message": message,
+                        })
+                    else:
+                        print(f"Error: {message}")
+                    sys.exit(1)
+                old_value = None
+                if use_json:
+                    from superlocalmemory.cli.json_output import json_print
+                    json_print("config", data={
+                        "key": key, "old_value": old_value,
+                        "new_value": result.get(field), "runtime": "daemon",
+                    })
+                else:
+                    print(f"{key}: applied to resident daemon -> {result.get(field)}")
+                return
+
         # Set via dot-notation (e.g. "evolution.enabled" -> cfg["evolution"]["enabled"])
         parts = key.split(".")
         node = cfg
@@ -1579,10 +1635,47 @@ def cmd_status(args: Namespace) -> None:
     from superlocalmemory.core.config import SLMConfig
 
     config = SLMConfig.load()
+    daemon_status = None
+    try:
+        from superlocalmemory.cli.daemon import (
+            daemon_request,
+            is_daemon_running,
+        )
+
+        if is_daemon_running():
+            candidate = daemon_request("GET", "/status")
+            if isinstance(candidate, dict) and candidate.get("profile"):
+                daemon_status = candidate
+    except Exception:
+        logger.debug(
+            "cmd_status: daemon runtime status unavailable; using offline view",
+            exc_info=True,
+        )
 
     if getattr(args, 'json', False):
         from superlocalmemory.cli.json_output import json_print
 
+        if daemon_status is not None:
+            data = {
+                "mode": str(daemon_status.get("mode", "unknown")).upper(),
+                "provider": daemon_status.get("provider", "none"),
+                "profile": daemon_status["profile"],
+                "base_dir": daemon_status.get("base_dir", str(config.base_dir)),
+                "db_path": daemon_status.get("db_path", str(config.db_path)),
+                "db_size_mb": float(daemon_status.get("db_size_mb", 0.0)),
+                "fact_count": int(daemon_status.get("fact_count", 0)),
+                "entity_count": int(daemon_status.get("entity_count", 0)),
+                "edge_count": int(daemon_status.get("edge_count", 0)),
+                "profile_generation": int(
+                    daemon_status.get("profile_generation", 0)
+                ),
+            }
+            json_print("status", data=data, next_actions=[
+                {"command": "slm health --json", "description": "Check math layer health"},
+                {"command": "slm list --json", "description": "List recent memories"},
+            ])
+            return
+
         # WP-02 D8: canonical key set — db_size_mb always present (0.0 if absent).
         db_size_mb = 0.0
         if config.db_path.exists():
@@ -1635,6 +1728,7 @@ def cmd_status(args: Namespace) -> None:
             "fact_count": fact_count,
             "entity_count": entity_count,
             "edge_count": edge_count,
+            "profile_generation": 0,
         }
         json_print("status", data=data, next_actions=[
             {"command": "slm health --json", "description": "Check math layer health"},
@@ -1645,6 +1739,10 @@ def cmd_status(args: Namespace) -> None:
     print("SuperLocalMemory V3")
     print(f"  Mode: {config.mode.value.upper()}")
     print(f"  Provider: {config.llm.provider or 'none'}")
+    print(
+        f"  Profile: "
+        f"{daemon_status.get('profile') if daemon_status else config.active_profile}"
+    )
     print(f"  Base dir: {config.base_dir}")
     print(f"  Database: {config.db_path}")
     if config.db_path.exists():
@@ -2522,6 +2620,42 @@ def cmd_dashboard(args: Namespace) -> None:
 # -- Profiles (supports --json) -------------------------------------------
 
 
+def _switch_profile_runtime(config, profile_name: str) -> dict:
+    """Switch through the resident daemon, or persist an offline fallback."""
+    from superlocalmemory.cli.daemon import daemon_request, is_daemon_running
+
+    if is_daemon_running():
+        result = daemon_request(
+            "POST",
+            f"/api/profiles/{profile_name}/switch",
+        )
+        if not result or not result.get("success"):
+            raise RuntimeError(
+                "resident daemon did not acknowledge the profile switch"
+            )
+        acknowledged = str(result.get("active_profile", ""))
+        if acknowledged != profile_name:
+            raise RuntimeError(
+                "resident daemon acknowledged a different active profile"
+            )
+        return {
+            "action": "switched",
+            "profile": acknowledged,
+            "generation": int(result.get("generation", 0)),
+            "runtime": "daemon",
+        }
+
+    from superlocalmemory.server.profile_runtime import persist_active_profile
+
+    persist_active_profile(profile_name)
+    config.active_profile = profile_name
+    return {
+        "action": "switched",
+        "profile": profile_name,
+        "runtime": "offline",
+    }
+
+
 def cmd_profile(args: Namespace) -> None:
     """Profile management (list, switch, create).
 
@@ -2532,7 +2666,7 @@ def cmd_profile(args: Namespace) -> None:
     from superlocalmemory.storage.database import DatabaseManager
     from superlocalmemory.storage import schema
     from superlocalmemory.server.routes.helpers import (
-        ensure_profile_in_json, set_active_profile_everywhere,
+        ensure_profile_in_json,
     )
 
     config = SLMConfig.load()
@@ -2552,10 +2686,25 @@ def cmd_profile(args: Namespace) -> None:
                            {"command": "slm profile switch <name> --json", "description": "Switch profile"},
                        ])
         elif args.action == "switch":
-            set_active_profile_everywhere(args.name)
-            config.active_profile = args.name
-            config.save()
-            json_print("profile", data={"action": "switched", "profile": args.name})
+            rows = db.execute(
+                "SELECT 1 FROM profiles WHERE profile_id = ?",
+                (args.name,),
+            )
+            if not rows:
+                json_print("profile", error={
+                    "code": "PROFILE_NOT_FOUND",
+                    "message": f"Profile '{args.name}' does not exist.",
+                })
+                sys.exit(1)
+            try:
+                result = _switch_profile_runtime(config, args.name)
+            except Exception as exc:
+                json_print("profile", error={
+                    "code": "PROFILE_SWITCH_FAILED",
+                    "message": str(exc),
+                })
+                sys.exit(1)
+            json_print("profile", data=result)
         elif args.action == "create":
             db.execute(
                 "INSERT OR IGNORE INTO profiles (profile_id, name) VALUES (?, ?)",
@@ -2576,10 +2725,21 @@ def cmd_profile(args: Namespace) -> None:
             d = dict(r)
             print(f"  - {d['profile_id']}: {d.get('name', '')}")
     elif args.action == "switch":
-        set_active_profile_everywhere(args.name)
-        config.active_profile = args.name
-        config.save()
-        print(f"Switched to profile: {args.name}")
+        rows = db.execute(
+            "SELECT 1 FROM profiles WHERE profile_id = ?",
+            (args.name,),
+        )
+        if not rows:
+            print(f"Profile '{args.name}' does not exist.", file=sys.stderr)
+            sys.exit(1)
+        try:
+            result = _switch_profile_runtime(config, args.name)
+        except Exception as exc:
+            print(f"Profile switch failed: {exc}", file=sys.stderr)
+            sys.exit(1)
+        generation = result.get("generation")
+        suffix = f" (generation {generation})" if generation is not None else ""
+        print(f"Switched to profile: {args.name}{suffix}")
     elif args.action == "create":
         db.execute(
             "INSERT OR IGNORE INTO profiles (profile_id, name) VALUES (?, ?)",

package/src/superlocalmemory/cli/setup_wizard.py

@@ -22,6 +22,7 @@ import subprocess
 import sys
 import time
 from pathlib import Path
+from typing import Any
 
 
 # ---------------------------------------------------------------------------
@@ -227,6 +228,46 @@ def _download_compressor(model_name: str) -> bool:
         return False
 
 
+# ---------------------------------------------------------------------------
+# Embedding-provider detection (v3.7.6 #72)
+# ---------------------------------------------------------------------------
+
+def _embedding_is_remote(config: Any) -> bool:
+    """True when embeddings come from a remote/OpenAI-compatible endpoint.
+
+    In that case the local sentence-transformers download (768d nomic) is
+    unnecessary and would only waste bandwidth/disk (#72). Detection is by
+    provider name or the presence of a configured HTTP endpoint.
+    """
+    emb = getattr(config, "embedding", None)
+    if emb is None:
+        return False
+    provider = (getattr(emb, "provider", "") or "").strip().lower()
+    endpoint = (
+        getattr(emb, "api_endpoint", "")
+        or getattr(emb, "base_url", "")
+        or ""
+    )
+    return provider in ("openai", "openai-compatible", "remote") or bool(endpoint)
+
+
+def _build_wizard_config(mode):
+    """Apply mode-owned presets without erasing existing user-owned blocks."""
+    from superlocalmemory.core.config import SLMConfig
+    from superlocalmemory.infra.data_root import state_path
+
+    template = SLMConfig.for_mode(mode)
+    if not state_path("config.json").exists():
+        return template
+    existing = SLMConfig.load()
+    existing.mode = mode
+    existing.llm = template.llm
+    existing.retrieval = template.retrieval
+    existing.math = template.math
+    existing.channel_weights = template.channel_weights
+    return existing
+
+
 # ---------------------------------------------------------------------------
 # Verification
 # ---------------------------------------------------------------------------
@@ -373,7 +414,7 @@ def run_wizard(auto: bool = False) -> None:
     from superlocalmemory.storage.models import Mode
 
     mode_map = {"a": Mode.A, "b": Mode.B, "c": Mode.C}
-    config = SLMConfig.for_mode(mode_map[choice])
+    config = _build_wizard_config(mode_map[choice])
 
     # -- Multi-scope (shared memory) opt-in — v3.6.15 --
     # OFF by default: your memories stay private to this profile (3.6.14 behaviour).
@@ -455,7 +496,17 @@ def run_wizard(auto: bool = False) -> None:
     print()
     print("─── Step 4/10: Download Embedding Model ───")
 
-    if not st_ok:
+    if _embedding_is_remote(config):
+        # v3.7.6 (#72): a remote/OpenAI-compatible embedding endpoint supplies
+        # its own vectors (any width, e.g. 1024d Qwen3-Embedding), so pulling the
+        # local 768d nomic model would be wasted bandwidth and could imply a
+        # dimension that does not match the endpoint.
+        emb = config.embedding
+        print("  ✓ Skipped — remote/OpenAI-compatible embedding endpoint configured")
+        print(f"    provider={getattr(emb, 'provider', '?')}, "
+              f"dimension={getattr(emb, 'dimension', '?')}")
+        print("    No local embedding model needed.")
+    elif not st_ok:
         print("  ⚠ Skipped (sentence-transformers not installed)")
         print("    Models will download on first use.")
     else:

package/src/superlocalmemory/core/backend_orchestrator.py

@@ -354,7 +354,12 @@ class BackendOrchestrator:
         try:
             from superlocalmemory.vector.lancedb_backend import LanceDBVectorBackend
             lance_path = self._data_dir / "lance"
-            self._lancedb = LanceDBVectorBackend(str(lance_path))
+            # v3.7.6 (#72): honor the configured embedding width instead of the
+            # hardcoded 768d, so custom endpoints (e.g. 1024d Qwen3-Embedding) work.
+            dimension = getattr(
+                getattr(self._config, "embedding", None), "dimension", None
+            )
+            self._lancedb = LanceDBVectorBackend(str(lance_path), dimension=dimension)
             self._update_status("lancedb", "not_initialized")
             logger.info("LanceDB initialized at %s", lance_path)
         except Exception as exc:

package/src/superlocalmemory/core/config.py

@@ -258,7 +258,7 @@ class RetrievalConfig:
     # Reranking (V3.3.2: ONNX backend enabled for all modes)
     # V3.4.2: Tested gte-reranker-modernbert-base (8K context) — REGRESSED
     # LoCoMo from 68.4% to 64.1%. Reverted to MiniLM-L-12-v2. The 512-token
-    # limit is acceptable because SLM's 6-channel retrieval pre-filters
+    # limit is acceptable because SLM's multi-producer retrieval pre-filters
     # relevant facts before reranking. See bench-v342-locomo.md.
     use_cross_encoder: bool = True
     cross_encoder_model: str = "cross-encoder/ms-marco-MiniLM-L-12-v2"

package/src/superlocalmemory/core/engine.py

@@ -51,8 +51,9 @@ class MemoryEngine:
     """Main orchestrator for the SuperLocalMemory V3 memory system.
 
     Wires encoding (fact extraction, entity resolution, graph building,
-    consolidation) with retrieval (4-channel search, RRF fusion,
-    reranking) and all supporting layers (trust, learning, compliance).
+    consolidation) with retrieval (five candidate producers -> RRF fusion,
+    reranking, entity-graph post-fusion enhancement) and all supporting
+    layers (trust, learning, compliance).
 
     Usage::
 

package/src/superlocalmemory/core/engine_wiring.py

@@ -500,7 +500,9 @@ def init_retrieval(
     trust_scorer: Any,
     vector_store: Any = None,
 ) -> Any:
-    """Create the RetrievalEngine with 6 channels. Returns it."""
+    """Create the RetrievalEngine — five candidate producers (semantic, BM25,
+    temporal, spreading_activation, hopfield) plus the entity graph used for
+    post-fusion score enhancement. Returns it."""
     from superlocalmemory.retrieval.engine import RetrievalEngine
     from superlocalmemory.retrieval.semantic_channel import SemanticChannel
     from superlocalmemory.retrieval.bm25_channel import BM25Channel

package/src/superlocalmemory/core/scale_engine.py

@@ -487,7 +487,15 @@ class ScaleEngineManager:
     def _real_backend_factory(self, cozo_dir: Path, lance_dir: Path) -> tuple[Any, Any]:
         from superlocalmemory.graph.cozo_backend import CozoDBGraphBackend
         from superlocalmemory.vector.lancedb_backend import LanceDBVectorBackend
-        return CozoDBGraphBackend(str(cozo_dir / "graph")), LanceDBVectorBackend(str(lance_dir))
+        # v3.7.6 (#72): promote the configured embedding width into the new
+        # LanceDB store so a 1024d (or other) custom endpoint survives promotion.
+        dimension = getattr(
+            getattr(self.config, "embedding", None), "dimension", None
+        )
+        return (
+            CozoDBGraphBackend(str(cozo_dir / "graph")),
+            LanceDBVectorBackend(str(lance_dir), dimension=dimension),
+        )
 
     def _readonly_connection(self) -> sqlite3.Connection:
         conn = sqlite3.connect(f"file:{self.db_path}?mode=ro", uri=True)

package/src/superlocalmemory/core/store_pipeline.py

@@ -310,7 +310,7 @@ def run_store(
     # that fact extraction may abstract away (dates, names, specifics).
     # This ensures BM25 and semantic search can always find the original text.
     # V3.3.12: Extract entities from verbatim content so entity channel + temporal
-    # channel can find it (was entities=[] which made 4/6 channels blind).
+    # channel can find it (was entities=[] which blinded the entity-graph and temporal signals).
     # V3.3.20: Stronger verbatim filter — skip greetings, filler, short phrases.
     # Verbatim facts with just "Hey! How are you?" dilute embeddings and add noise.
     _MIN_VERBATIM_WORDS = 8

package/src/superlocalmemory/hooks/before_web_hook.py

@@ -17,7 +17,7 @@ the local memories before committing to the external call.
 
 PERFORMANCE
 ===========
-Cost: ~500-800ms warm (full 4-channel recall via SLM daemon). Fires only on
+Cost: ~500-800ms warm (full recall via SLM daemon). Fires only on
 WebSearch and WebFetch (5-20× per typical session), so per-session overhead
 is ~5-15s in exchange for grounded answers. NOT suitable for UserPromptSubmit
 (too frequent — would be a perf disaster).

package/src/superlocalmemory/hooks/claude_code_hooks.py

@@ -267,7 +267,7 @@ def _hook_definitions(include_gate: bool = False) -> dict[str, list]:
     }
 
     # v3.4.43 — default PreToolUse entry: pre-web recall on WebSearch/WebFetch.
-    # Fires `slm hook before_web` which runs a 4-channel recall on the search
+    # Fires `slm hook before_web` which runs a recall on the search
     # query/URL and injects results as a system-reminder BEFORE the web call.
     # Encourages Claude to consider local memories before paying for new web
     # research. Independent of `include_gate` — this is value-add, not gating.

package/src/superlocalmemory/infra/auth_middleware.py

@@ -1,12 +1,12 @@
 # Copyright (c) 2026 Varun Pratap Bhardwaj / Qualixar
 # Licensed under AGPL-3.0-or-later - see LICENSE file
 # Part of SuperLocalMemory V3 | https://qualixar.com | https://varunpratap.com
-"""Opt-in API-key authentication middleware.
+"""API-key primitives used by the unified identity boundary.
 
-When ``~/.superlocalmemory/api_key`` exists, write endpoints require the
-``X-SLM-API-Key`` header.  Read endpoints remain open for backward
-compatibility.  If the key file is absent auth is completely disabled --
-all requests pass.
+A configured key authorizes remote callers that present ``X-SLM-API-Key``.
+The unified daemon separately trusts exact-process capabilities, the
+same-origin dashboard install token, and uncredentialed loopback peers as the
+local OS-user boundary. Read endpoints remain open for backward compatibility.
 
 V3 change: base directory moved from ``~/.claude-memory/`` to
 ``~/.superlocalmemory/``.