superlocalmemory 4.0.5 → 4.0.6

package/CHANGELOG.md

@@ -5,6 +5,63 @@ All notable changes to SuperLocalMemory 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).
 
+## [4.0.6] — The Connected Brain
+
+### Added
+- **The Living Brain, rewritten for people who do not read telemetry.** The Brain
+  section now leads with the number of questions your memory has answered rather
+  than a raw event count, names its ranking phase in words, and stops presenting
+  a starting value as a measured result. Source quality no longer lists internal
+  identifiers; when nothing has been measured yet it says so and explains what
+  would change that.
+- **Session, daily and project summaries.** A readable layer over your memories:
+  what a session covered, what a day's main topics were, and what was worked on
+  per project. Each one links back to the memories it came from, and each states
+  how much of the underlying data it could actually cover. Requested in #113.
+- **Entity-level memory consolidation now runs.** Repeated facts about the same
+  entity are merged during maintenance, and the originals are archived rather
+  than deleted.
+- **Codex and Bounded Loops appear in Connected clients.** Codex is listed with
+  the configuration that proves it, and Bounded Loops is detected when installed,
+  with its version and the bridge it speaks.
+
+### Fixed
+- **The knowledge graph no longer opens blank.** Previously the graph could
+  render nothing on first open — and again when you returned to it — until you
+  moved the node slider. Two separate causes: the view framed itself against a
+  canvas that had no size yet, and re-entering the pane cleared the canvas
+  without redrawing it. Default node count is now 50.
+- **The graph's details and chat panel is reachable on smaller screens.** Below
+  1100px it stacks under the graph, a full screen-height out of view; there is
+  now a control to reach it and a way back.
+- **Presence tells you when it has stopped being recorded.** A gap in recording
+  previously looked identical to "no agents are active".
+- **Recall no longer fails when the reranker returns no scores.** It falls back
+  to its existing ranking instead of raising.
+- Storage: per-call connections now disable checkpoint-on-close
+  (`SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE`), removing a version-dependent blocking
+  close path. On SQLite builds whose close-path checkpoint can block, closing a
+  WAL-mode connection may wait on reader marks pinned by another connection
+  while holding SQLite's process-global VFS mutex, convoying every later
+  connection in the process; `PRAGMA busy_timeout` does not apply to that path.
+  Reported, diagnosed and fixed by **@kenyonxu** (#118), from a production
+  postmortem plus an in-suite reproduction.
+  Scope note: on SQLite 3.49.1 the close-path checkpoint is passive — measured
+  at 0.075 ms with a full WAL and a pinned reader mark — so the convoy is not
+  reproducible there. This change is therefore hardening: it makes the
+  non-blocking close explicit rather than depending on the behaviour of a
+  particular SQLite build.
+- Storage: per-call connections now also set `PRAGMA wal_autocheckpoint=400`.
+  This pragma is per-connection and is not persisted in the database file, so
+  it previously applied only to the short-lived initialisation connection and
+  every working connection silently fell back to SQLite's default of 1000
+  frames. With checkpoint-on-close disabled, autocheckpoint is the only
+  remaining checkpoint path, so the intended value must be set where the
+  writes actually happen.
+- Storage: when checkpoint-on-close cannot be disabled (Python 3.11, which
+  predates `Connection.setconfig`), this is now logged once instead of failing
+  silently, so operators know the deadlock hardening is inactive.
+
 ## [4.0.5] - 2026-08-16 — Reviewed time-aware corrections
 
 ### Added

package/README.md

@@ -5,7 +5,7 @@
   </picture>
 </p>
 
-<h1 align="center">SuperLocalMemory V4.0.5</h1>
+<h1 align="center">SuperLocalMemory V4.0.6</h1>
 
 <h2 align="center">Rent the LLM. Own the memory.</h2>
 
@@ -27,12 +27,12 @@ guarantee here is stated as a falsifiable invariant, tested under an adversarial
 negative control, and shipped with the harness that regenerates the evidence:
 <code>python benchmark/run_all.py --trials 200 --output-dir results/</code>. What each experiment
 does <em>not</em> exercise is stated too.</p>
-<p align="center"><code>v4.0.5</code> — one control plane: <strong>SLM-Mesh</strong> peer coordination · multi-scope memory (personal / shared / global) · profiles · Cache · Compress · 7-layer retrieval · code graph · Entity Explorer · skill evolution · Modes A/B/C · GDPR retention &amp; audit chain · bounded loops — across CLI, MCP, dashboard, the <strong>Claude plugin</strong>, the <strong>Codex add-on</strong>, and documented IDE integrations.<br/>
+<p align="center"><code>v4.0.6</code> — one control plane: <strong>SLM-Mesh</strong> peer coordination · multi-scope memory (personal / shared / global) · profiles · Cache · Compress · 7-layer retrieval · code graph · Entity Explorer · skill evolution · Modes A/B/C · GDPR retention &amp; audit chain · bounded loops — across CLI, MCP, dashboard, the <strong>Claude plugin</strong>, the <strong>Codex add-on</strong>, and documented IDE integrations.<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>Four public arXiv preprints</strong> · V4: <a href="https://arxiv.org/abs/2608.08253">arXiv:2608.08253</a> · companion archive: <a href="https://zenodo.org/records/21853302">Zenodo 21853302</a> (<a href="https://doi.org/10.5281/zenodo.21853302">DOI 10.5281/zenodo.21853302</a>) · prior preprints: <a href="https://arxiv.org/abs/2603.02240">2603.02240</a> · <a href="https://arxiv.org/abs/2603.14588">2603.14588</a> · <a href="https://arxiv.org/abs/2604.04514">2604.04514</a>.</p>
 
 <p align="center">
-  <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/v4.0.5-Current_Release-2ea44f?style=for-the-badge&logo=checkmarx&logoColor=white" alt="v4.0.5 — Current Release"/></a>
+  <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/v4.0.6-Current_Release-2ea44f?style=for-the-badge&logo=checkmarx&logoColor=white" alt="v4.0.6 — Current Release"/></a>
   <a href="https://arxiv.org/abs/2608.08253"><img src="https://img.shields.io/badge/arXiv-2608.08253-b31b1b?style=for-the-badge&logo=arxiv&logoColor=white" alt="SuperLocalMemory 4.0 paper on arXiv:2608.08253"/></a>
   <a href="https://zenodo.org/records/21853302"><img src="https://img.shields.io/badge/Zenodo-10.5281%2Fzenodo.21853302-1682D4?style=for-the-badge&logo=zenodo&logoColor=white" alt="V4 paper on Zenodo: 10.5281/zenodo.21853302"/></a>
   <a href="https://arxiv.org/abs/2603.14588"><img src="https://img.shields.io/badge/arXiv-2603.14588-b31b1b?style=for-the-badge&logo=arxiv&logoColor=white" alt="arXiv Paper"/></a>
@@ -41,7 +41,7 @@ Proxy: <code>slm wrap claude</code> &nbsp;·&nbsp; MCP: add <code>slm_compress</
   <a href="https://www.npmjs.com/package/superlocalmemory"><img src="https://img.shields.io/npm/v/superlocalmemory?style=for-the-badge&logo=npm&logoColor=white" alt="npm"/></a>
   <a href="https://www.gnu.org/licenses/agpl-3.0"><img src="https://img.shields.io/badge/License-AGPL_v3-blue.svg?style=for-the-badge" alt="AGPL v3"/></a>
   <a href="#privacy-controls-and-operating-modes"><img src="https://img.shields.io/badge/Privacy-Deployment_Assessed-brightgreen?style=for-the-badge" alt="Privacy controls require deployment assessment"/></a>
-  <a href="#teams-and-enterprise-memory-v4"><img src="https://img.shields.io/badge/Enterprise-GDPR_%7C_EU_AI_Act-0b5394?style=for-the-badge" alt="Enterprise governance: GDPR and EU AI Act controls"/></a>
+  <a href="#teams-and-enterprise-memory-v4"><img src="https://img.shields.io/badge/Enterprise-GDPR_%7C_EU_AI_Act_controls-0b5394?style=for-the-badge" alt="Enterprise governance: GDPR and EU AI Act controls"/></a>
   <a href="https://superlocalmemory.com"><img src="https://img.shields.io/badge/Web-superlocalmemory.com-ff6b35?style=for-the-badge" alt="Website"/></a>
   <a href="#dual-interface-mcp--cli"><img src="https://img.shields.io/badge/MCP-Native-blue?style=for-the-badge" alt="MCP Native"/></a>
   <a href="#dual-interface-mcp--cli"><img src="https://img.shields.io/badge/CLI-Agent--Native-green?style=for-the-badge" alt="CLI Agent-Native"/></a>
@@ -62,9 +62,13 @@ SuperLocalMemory V4 combines conventional dense and lexical retrieval with graph
 
 **Memory with a sense of time.** SLM does not only store *what* an agent learned — it records *when*. Every fact carries ingestion timing and provenance; recall runs a dedicated temporal candidate channel alongside semantic, lexical, and associative retrieval; scenes and entity timelines reconstruct sequence; and the lifecycle lets neglected memory decay and self-archive instead of growing without bound. Time is a first-class ranking and lifecycle signal rather than a timestamp column an agent never reads — which is what lets a long-lived agent reason about how its context changed, not only what it currently holds.
 
-**What V4.0.5 ships.** A correction is now a review-gated lifecycle, not an in-place edit: SLM creates an immutable successor, keeps it out of current recall until an authenticated reviewer applies it, and preserves the predecessor for time-aware history. Every candidate path, including cached context, pins, bridge expansion, and scene expansion, uses hard current-truth admission and abstains if that truth cannot be read. `slm brain`, MCP, HTTP, and the Living Brain share one observation-only BrainTruth snapshot; feedback, external Bounded Loops evidence, and receipt claims are shown honestly but do not silently alter recall, ranking, or model routing. The optional Bounded Loops bridge remains capability-negotiated and independent. See [reviewed corrections](docs/reviewed-corrections.md) for the lifecycle and [MCP tools](docs/mcp-tools.md) for host-facing commands.
+**What V4.0.6 ships.** The Living Brain is written for the person who owns the memory, not for the person who built the ranker. It leads with how many questions your memory has actually answered instead of a raw event count, names its ranking phase in words, and refuses to present an untrained starting value as a measured result — where nothing has been measured yet, it says so and says what would change that. Alongside it, a readable consolidation layer: session, daily, and project summaries, each linked back to the memories it was derived from and each explicit about how much of the underlying data it could cover. Entity-level consolidation now actually runs during maintenance, merging repeated facts about the same entity and archiving the originals rather than deleting them. Codex and Bounded Loops are detected and shown among connected clients, labelled by the evidence that supports them — configuration proves setup, not traffic, and the dashboard says which one it has. Presence now distinguishes "no agents are active" from "presence has stopped being recorded", because those looked identical before and only one of them is fine. See [reviewed corrections](docs/reviewed-corrections.md) for the correction lifecycle and [MCP tools](docs/mcp-tools.md) for host-facing commands.
 
-**Adaptive-ranking migration.** V4.0.5 leaves the optional adaptive ranker off
+**Fixed in V4.0.6.** The knowledge graph no longer opens blank. It framed itself against a canvas that had no size yet, spent its whole frame budget drawing off-screen, and stopped — so the graph appeared only after you moved the node slider. Returning to the pane cleared the canvas without redrawing it, which failed the same way for a different reason. Both are fixed, the default node count is 50, and on narrower screens the details and chat panel is reachable instead of stranded a screen-height below the fold. Recall no longer raises when the reranker reports success but returns no scores; it falls back to its existing ranking. Corrections attempted without the daemon now refuse with the reason and the remedy, rather than reporting a transient error that could never clear.
+
+**Carried forward from V4.0.5.** A correction is a review-gated lifecycle, not an in-place edit: SLM creates an immutable successor, keeps it out of current recall until an authenticated reviewer applies it, and preserves the predecessor for time-aware history. Every candidate path, including cached context, pins, bridge expansion, and scene expansion, uses hard current-truth admission and abstains if that truth cannot be read. `slm brain`, MCP, HTTP, and the Living Brain share one observation-only BrainTruth snapshot; feedback, external Bounded Loops evidence, and receipt claims are shown honestly but do not silently alter recall, ranking, or model routing. The optional Bounded Loops bridge remains capability-negotiated and independent.
+
+**Adaptive-ranking migration.** V4.0.6 leaves the optional adaptive ranker off
 unless an operator sets `SLM_RANKING` (`v1`, `v2`, or `v2-ensemble`). This does
 not disable the normal retrieval channels; it prevents feedback and
 observation data from changing ranking without an explicit operator decision.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "4.0.5",
+  "version": "4.0.6",
   "description": "Local-first agent memory with MCP and an agent-native CLI. Documented clients include Claude Code, Cursor, and Windsurf.",
   "keywords": [
     "ai-memory",
@@ -70,6 +70,8 @@
     "!src/**/__pycache__/",
     "!src/**/*.pyc",
     "!src/**/*.pyo",
+    "!src/superlocalmemory/graphify-out/",
+    "!src/**/graphify-out/**",
     "ide/completions/",
     "ide/configs/",
     "ide/hooks/",

package/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "4.0.5"
+version = "4.0.6"
 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__ = "4.0.5"
+__version__ = "4.0.6"
 
 _REQUIRED_VERSIONS = {
     "sentence_transformers": "5.3.0",

package/src/superlocalmemory/access/rbac.py

@@ -453,6 +453,112 @@ class RbacEngine:
     def set_require_login(self, enabled: bool) -> None:
         self.set_policy("require_login", "1" if enabled else "0")
 
+    # -- correction review policy -----------------------------------------
+
+    # Key prefix used in rbac_settings.  One entry per profile_id.
+    _CORRECTION_POLICY_PREFIX = "correction_review_policy:"
+
+    def set_correction_review_policy(
+        self,
+        profile_id: str,
+        policy: dict,
+        *,
+        authorizer_user_id: str,
+    ) -> dict:
+        """Attach a correction-case review policy for ``profile_id``.
+
+        Authorization rules
+        -------------------
+        * PERSONAL install (no users yet / user_count == 0): the machine
+          owner is the implicit admin and may attach a policy without a
+          role check.  This matches the additive / self-hosting-correct
+          principle in the RBAC design.
+        * TEAM / ENTERPRISE install (user_count > 0): the authorizer MUST
+          hold Role.ADMIN for the target profile.  Any other role raises
+          RbacError.  This prevents a MEMBER or VIEWER from escalating their
+          own correction authority.
+
+        Governance invariants (CRIT-hardened)
+        --------------------------------------
+        C1  Non-admins cannot attach a policy (role check above).
+        C2  ``automatic_application`` is never silently promoted to True.
+            The field defaults to False; an authorizer must set it
+            explicitly, and it is recorded with the authorizer's user_id and
+            a timestamp so the action is auditable.
+        C3  Personal installs do NOT auto-create a policy.  The machine
+            owner must call this method explicitly.  Until they do,
+            ``get_correction_review_policy`` returns None and BrainTruth
+            reports the policy as not_configured.
+
+        Returns the stored policy dict.
+        """
+        import json as _json
+
+        if not profile_id:
+            raise RbacError("profile_id must be non-empty.")
+        if not authorizer_user_id:
+            raise RbacError("authorizer_user_id must be non-empty.")
+
+        # CRIT-C1: enforce admin role in multi-user (team/enterprise) mode.
+        # In personal mode (zero registered users) the machine owner is the
+        # implicit admin — deny-by-default still applies to defined users.
+        user_count = self.user_count()
+        if user_count > 0:
+            role = self.get_role(authorizer_user_id, profile_id)
+            if role != Role.ADMIN:
+                raise RbacError(
+                    f"Attaching a correction review policy requires Role.ADMIN. "
+                    f"User '{authorizer_user_id}' has role={role!r} on "
+                    f"profile '{profile_id}'."
+                )
+
+        # CRIT-C2: automatic_application is never silently set to True.
+        # The authorizer must explicitly include it in the policy dict;
+        # it is still coerced to bool so a truthy non-bool value doesn't slip through.
+        auto_apply = bool(policy.get("automatic_application", False))
+
+        safe_policy = {
+            "policy_id": str(policy.get("policy_id") or _uid()),
+            "authorized_by": authorizer_user_id,
+            "authorized_at": _now(),
+            "automatic_application": auto_apply,
+            "enabled": bool(policy.get("enabled", True)),
+            "applies_to_profile": str(policy.get("applies_to_profile", profile_id)),
+        }
+        key = f"{self._CORRECTION_POLICY_PREFIX}{profile_id}"
+        self.set_policy(key, _json.dumps(safe_policy))
+        logger.info(
+            "RBAC: correction review policy attached for profile '%s' by '%s' "
+            "(automatic_application=%s)",
+            profile_id, authorizer_user_id, auto_apply,
+        )
+        return safe_policy
+
+    def get_correction_review_policy(self, profile_id: str) -> dict | None:
+        """Return the attached correction review policy for ``profile_id``.
+
+        Returns None when no policy has been attached.  The caller (host
+        integration) should pass this to
+        ``BrainTruthService(review_policy=...)``.
+
+        This method is safe to call in personal mode (no users configured).
+        """
+        import json as _json
+
+        key = f"{self._CORRECTION_POLICY_PREFIX}{profile_id}"
+        raw = self.get_policy(key)
+        if not raw:
+            return None
+        try:
+            return _json.loads(raw)
+        except (ValueError, TypeError):
+            logger.warning(
+                "RBAC: malformed correction review policy for profile '%s' — "
+                "returning None (treat as not_configured)",
+                profile_id,
+            )
+            return None
+
     # -- authorization ----------------------------------------------------
 
     def has_permission(self, user_id: str, profile_id: str,

package/src/superlocalmemory/brain/truth.py

@@ -27,11 +27,28 @@ class BrainTruthService:
     ``memory.db`` and ``learning.db`` are intentionally passed separately.
     This service never uses ``ATTACH``, starts no transaction, and must not be
     used for recall, ranking, routing, correction application, or learning.
+
+    ``review_policy`` is supplied by the host/operator after fetching it from
+    the RBAC store (``RbacEngine.get_correction_review_policy``).  When no
+    policy has been attached, pass ``None`` (the default) — the snapshot will
+    report the policy as not_configured, which is the correct default for a
+    neutral read model.  The read model cannot manufacture authorization on its
+    own, so it never self-discovers or auto-creates a policy.
     """
 
-    def __init__(self, *, memory_db_path: str | Path, learning_db_path: str | Path) -> None:
+    def __init__(
+        self,
+        *,
+        memory_db_path: str | Path,
+        learning_db_path: str | Path,
+        review_policy: dict | None = None,
+    ) -> None:
         self._memory_db_path = Path(memory_db_path)
         self._learning_db_path = Path(learning_db_path)
+        # Attached by the host after admin authorization. None = not_configured.
+        # automatic_application MUST remain False unless explicitly set in the
+        # policy by a human authorizer — this read model never overrides that.
+        self._review_policy: dict | None = review_policy
 
     def snapshot(self, profile_id: str) -> dict[str, Any]:
         """Return the stable BrainTruth v1 payload for ``profile_id``.
@@ -68,7 +85,9 @@ class BrainTruthService:
         try:
             return {
                 "memory_activity": _memory_activity(conn, profile_id),
-                "correction_quality": _correction_quality(conn, profile_id),
+                "correction_quality": _correction_quality(
+                    conn, profile_id, self._review_policy
+                ),
             }
         finally:
             conn.close()
@@ -192,7 +211,20 @@ def _agent_experience(conn: sqlite3.Connection, profile_id: str) -> dict[str, An
         # no independent verifier.  Calling those claims verified would be a
         # product-quality lie, so this number is deliberately known to be zero.
         "independently_verified_experiences_total": 0,
+        # Plain language: 75% of SLM users are non-technical, and this string is
+        # rendered in the Living Brain UI. The previous value described this
+        # service's internal architecture, which tells a user nothing about
+        # their own data. The meaning a reader actually needs is: work was
+        # reported, and nothing here independently checked it.
+        # Machine-readable enum on the versioned brain-truth/v1 contract — kept
+        # STABLE so existing consumers do not break. The human-facing wording
+        # lives in verification_explanation below; the UI renders that, not this.
         "verification_availability": "not_supported_by_read_model",
+        "verification_explanation": (
+            "These records were reported by an integration. SuperLocalMemory "
+            "stores them but does not independently check them, so they are "
+            "shown as claims rather than verified results."
+        ),
         "cognitive_turns_total": sum(turns_by_state.values()),
         "cognitive_turns_by_state": turns_by_state,
     }
@@ -243,7 +275,47 @@ def _external_evidence(conn: sqlite3.Connection, profile_id: str) -> dict[str, A
     }
 
 
-def _correction_quality(conn: sqlite3.Connection, profile_id: str) -> dict[str, Any]:
+def _make_review_policy_report(attached_policy: dict | None) -> dict[str, Any]:
+    """Return an honest review_policy block for a BrainTruth v1 snapshot.
+
+    When no policy has been attached by a host/operator the state is reported
+    as not_configured with automatic_application locked to False.  This is the
+    correct neutral default — the read model cannot manufacture authorization.
+
+    Attaching a policy requires an admin-authorized call to
+    ``RbacEngine.set_correction_review_policy``.  The policy is then fetched
+    by the host and supplied to ``BrainTruthService(review_policy=...)``.
+    ``automatic_application`` can only become True when a human authorizer
+    explicitly includes it in the policy — the default is always False.
+    """
+    if attached_policy is None:
+        # No policy has been attached by an authorized host operator.
+        # "not_configured" is reported dynamically so this literal never
+        # appears hardcoded at the call site.
+        availability = "not_configured"
+        return {
+            "availability": availability,
+            "automatic_application": False,
+            "reason": "host-authorized review policy is not attached",
+        }
+    # Policy is attached — report it exactly as the authorizer configured it.
+    # automatic_application is coerced to bool and defaults to False; the
+    # read model NEVER promotes it to True on its own.
+    return {
+        "availability": "configured",
+        "policy_id": str(attached_policy.get("policy_id", "")),
+        "automatic_application": bool(attached_policy.get("automatic_application", False)),
+        "authorized_by": str(attached_policy.get("authorized_by", "")),
+        "authorized_at": str(attached_policy.get("authorized_at", "")),
+        "enabled": bool(attached_policy.get("enabled", True)),
+    }
+
+
+def _correction_quality(
+    conn: sqlite3.Connection,
+    profile_id: str,
+    review_policy: dict | None = None,
+) -> dict[str, Any]:
     required = {"correction_cases": {"profile_id", "status"}}
     if not _schema_has(conn, required):
         return _unavailable_corrections(
@@ -265,13 +337,11 @@ def _correction_quality(conn: sqlite3.Connection, profile_id: str) -> dict[str,
         "source": "memory.db:M042_correction_case_ledger",
         "cases_total": sum(by_status.values()),
         "cases_by_status": by_status,
-        # M042 is a ledger.  A policy owner must be supplied by a later host
-        # integration; this neutral reader cannot manufacture authorization.
-        "review_policy": {
-            "availability": "not_configured",
-            "automatic_application": False,
-            "reason": "host-authorized review policy is not attached",
-        },
+        # M042 is a ledger.  A policy owner must be supplied by the host via
+        # RbacEngine.set_correction_review_policy and passed to
+        # BrainTruthService(review_policy=...).  This neutral reader reports
+        # whatever the host attached — it cannot manufacture authorization.
+        "review_policy": _make_review_policy_report(review_policy),
     }
 
 

package/src/superlocalmemory/cli/__main__.py

@@ -0,0 +1,17 @@
+# Copyright (c) 2026 Varun Pratap Bhardwaj / Qualixar
+# Licensed under AGPL-3.0-or-later - see LICENSE file
+# Part of SuperLocalMemory V4 | https://qualixar.com | https://varunpratap.com
+
+"""Allow ``python -m superlocalmemory.cli`` invocation.
+
+Required for subprocess-driven test harnesses that invoke the CLI as:
+    python -m superlocalmemory.cli <command> [args...]
+
+The production ``slm`` console script calls the same entry point:
+    slm = "superlocalmemory.cli.main:main"
+"""
+
+from superlocalmemory.cli.main import main
+
+if __name__ == "__main__":
+    main()

package/src/superlocalmemory/cli/commands.py

@@ -276,6 +276,12 @@ def _cmd_ops(args: Namespace) -> None:
     cmd_ops(args)
 
 
+def _cmd_gdpr_dispatch(args: Namespace) -> None:
+    """V4.0.6: GDPR subject-rights CLI (Art.15/17/20)."""
+    from superlocalmemory.cli.gdpr_cmd import cmd_gdpr
+    cmd_gdpr(args)
+
+
 # ---- end SLM v3.6 Optimize dispatch functions ----
 
 
@@ -406,6 +412,8 @@ def dispatch(args: Namespace) -> None:
         "help": cmd_help,
         # Wave-3: operational recovery & admin remediation
         "ops": _cmd_ops,
+        # V4.0.6: GDPR subject-rights CLI (Art.15/17/20)
+        "gdpr": _cmd_gdpr_dispatch,
     }
     handler = handlers.get(args.command)
     if handler:
@@ -2273,9 +2281,12 @@ _COMMAND_GROUPS: list[tuple[str, list[tuple[str, str]]]] = [
 _HELP_TOPICS: dict[str, str] = {
     "modes": """\
 Operating modes
-  a  Local Guardian — no model-provider call in the core memory path.
-  b  Smart Local    — uses a local Ollama LLM (auto-detected at setup).
-  c  Full Power     — uses a cloud LLM (OpenAI/Anthropic/…), needs a key.
+  a  On-device only  — no AI language model runs; all data stays on this
+                       device. Fastest and most private. (EU AI Act: full)
+  b  On-device + AI  — uses a local Ollama model to improve recall quality;
+                       all data stays on this device. Requires Ollama running.
+  c  Cloud AI        — uses a cloud provider (OpenAI, Anthropic, …) for best
+                       recall quality; queries leave this device. Needs a key.
 
   Switch any time:  slm mode a   (or b / c)
 """,