superlocalmemory 3.8.11 → 3.8.13

package/CHANGELOG.md

@@ -5,6 +5,91 @@ 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).
 
+## [3.8.13] - 2026-08-03 — Stale-process detection
+
+### Fixed
+- A running process can now tell when it is serving superseded code. Python
+  imports a module once, so `__version__` is frozen at process start and
+  upgrading the package underneath a long-lived `slm mcp` server changes
+  nothing for that server — it keeps serving the code it read at startup,
+  indefinitely. Nothing detected this. The stale process did not error; it
+  returned confident, plausible, wrong answers and reported a
+  `serverInfo.version` matching the code it had loaded, which is
+  self-consistent and therefore useless as a staleness signal. During the
+  3.8.12 work one machine had eighteen `slm mcp` processes alive at once
+  spanning four days and two releases, and one of them made issue #106 look
+  unfixed across two debugging sessions. (#107)
+
+  The existing process reaper does not cover this: it kills *orphans*, whose
+  parent has died. A server whose IDE is still running is never an orphan.
+
+  Staleness is now reported by `slm doctor` (as a warning, with a restart
+  hint), on the loopback `/health` payload as `version_integrity`, and in the
+  `slm mcp` startup log. The MCP path logs to stderr only — that transport is
+  JSON-RPC over stdio, where a printed warning would corrupt the protocol and
+  turn a cosmetic problem into a dead session.
+
+  Running *ahead* of the installed distribution — normal for an editable
+  checkout — is deliberately reported separately and does not warn. A warning
+  that fires on every developer machine is one everybody learns to ignore, and
+  then it goes unread on the day it matters. Every failure path resolves to
+  `unknown` rather than to a false `current`.
+
+## [3.8.12] - 2026-08-03 — Canonical learning signals, clock-independent daemon identity, remote reranker
+
+### Fixed
+- Explicit feedback now reaches the store every consumer actually reads.
+  Three tables carry a "feedback" name: `feedback_records` (memory.db, read by
+  nothing), `learning_feedback` (the pre-v3.4.22 legacy table that
+  `legacy_migration` copies forward), and `learning_signals` +
+  `learning_features` (canonical — read by the dashboard Living Brain panel,
+  the ranker-phase card, and the LightGBM retrainer). The 3.8.11 fix wrote to
+  the legacy table, so it moved a counter nothing consumes. `record_explicit`
+  now writes the legacy row and the canonical signal/feature pair in one
+  transaction, flagged `is_synthetic=1` so the LightGBM trainer's
+  `WHERE is_synthetic=0` filter excludes them from training. (#106)
+- The recall phase gate and the dashboard no longer read different tables.
+  The gate counted `learning_feedback` while every user-visible surface counted
+  `learning_signals`, so the phase a user was shown and the phase that actually
+  ranked their results could disagree without limit. Both now resolve from the
+  canonical store, with thresholds imported from `learning.ranker` rather than
+  duplicated as literals — duplicated literals are how the two surfaces drifted
+  apart in the first place. (#106)
+- `report_feedback` no longer reports success for a write that did not happen.
+  It fell back to the `feedback_records` count when the canonical read failed,
+  and that table increments on every call regardless, so total write failure was
+  indistinguishable from success. The cross-store fallback is removed; a failed
+  durable write returns `success: false` with `durable: false`. (#106)
+- The CLI no longer disowns a healthy daemon after a few minutes on WSL2.
+  psutil derives `create_time` as boot time plus start ticks and re-reads
+  `/proc/stat` `btime` on every call; WSL2 resyncs its VM clock mid-session, so
+  `btime` moves and every process's computed creation time moves with it,
+  retroactively. The recorded value stopped matching the same live PID while
+  `/health` still answered in milliseconds. Ownership now compares a
+  clock-independent process token — `boot_id` plus raw start ticks on
+  Linux/WSL2, psutil's monotonic creation time elsewhere — as exact equality,
+  with no tolerance constant left to silently expire. PID-reuse protection is
+  strengthened: a creation-time mismatch no longer condemns a process outright,
+  it falls through to cryptographic identity proof over loopback. (#104)
+- `DAEMON_UNAVAILABLE` now names one of eight specific reasons with an
+  actionable hint instead of "owned daemon is unavailable; retry later". (#104)
+- `slm setup` no longer wipes the `retrieval` config block on every re-run.
+
+### Added
+- Remote and custom reranker endpoints, mirroring the remote embedding support
+  added in v3.4.24. Setting `cross_encoder_backend` to `openai` (or `remote`)
+  with a `cross_encoder_endpoint` routes reranking to an OpenAI-compatible
+  `/v1/rerank` service instead of the local subprocess worker. The built-in
+  cross-encoder is English-only, silently degrading recall for non-English
+  users; this lets them bring a multilingual model. Failure degrades to fusion
+  order with `applied=False` and an error log — never a silent fall back to the
+  local English model, which would recreate the very problem this solves. The
+  new outbound HTTP surface enforces an http/https allow-list, rejects 3xx and
+  does not follow redirects, refuses credentials embedded in the URL, and
+  bounds response reads at 8 MB. (#105)
+- `cross_encoder_endpoint` is now a real config field. It was previously
+  accepted and silently ignored — the remaining half of #103.
+
 ## [3.8.11] - 2026-08-02 — Learning-signal integrity and honest reranker diagnostics
 
 ### Fixed

package/README.md

@@ -5,15 +5,15 @@
   </picture>
 </p>
 
-<h1 align="center">SuperLocalMemory V3.8.11</h1>
+<h1 align="center">SuperLocalMemory V3.8.13</h1>
 <p align="center"><strong>Enterprise-grade, local-first memory for AI agents and teams.</strong><br/>
 <em>A persistent, auditable long-term brain for your agents that runs on your own infrastructure — with multi-workspace isolation, role-based access, and GDPR + EU AI Act governance controls built in.</em></p>
-<p align="center"><code>v3.8.11</code> — one control plane: auditable retrieval · multi-scope memory (personal / shared / global) · Cache · Compress · trusted-peer Mesh · 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>v3.8.13</code> — one control plane: auditable retrieval · multi-scope memory (personal / shared / global) · Cache · Compress · trusted-peer Mesh · 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>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>
 
 <p align="center">
-  <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/v3.8.11-Current_Release-2ea44f?style=for-the-badge&logo=checkmarx&logoColor=white" alt="v3.8.11 — Current Release"/></a>
+  <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/v3.8.13-Current_Release-2ea44f?style=for-the-badge&logo=checkmarx&logoColor=white" alt="v3.8.13 — Current Release"/></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>
   <a href="#three-surfaces-proxy--mcp-tools--skill"><img src="https://img.shields.io/badge/Proxy_|_MCP_|_Skill-22c55e?style=for-the-badge" alt="Three Surfaces: Proxy, MCP Tools, Skill"/></a>
   <a href="https://pypi.org/project/superlocalmemory/"><img src="https://img.shields.io/pypi/v/superlocalmemory?style=for-the-badge&logo=pypi&logoColor=white" alt="PyPI"/></a>
@@ -294,6 +294,10 @@ quality must be evaluated for the target client and workload; V3.8.0 publishes n
 
 **Multilingual models:** configure an OpenAI-compatible embedding endpoint such as Ollama, vLLM, LiteLLM, `bge-m3`, `multilingual-e5`, or `Qwen3-Embedding`. Language coverage and retrieval quality depend on the selected model and should be evaluated for the deployment corpus.
 
+<a id="remote-embedding-and-rerank-endpoints"></a>
+
+**Remote embedding + rerank endpoints.** The bundled reranker `cross-encoder/ms-marco-MiniLM-L-12-v2` is **English-only**, so a Chinese, Japanese, or Arabic corpus is scored by a model that cannot read it. Set `retrieval.cross_encoder_backend: "openai"` plus `retrieval.cross_encoder_endpoint` to route reranking to any Cohere-shaped `POST /v1/rerank` service — llama-server, text-embeddings-inference, Infinity — running a multilingual model such as `BAAI/bge-reranker-v2-m3` (v3.8.12, [#105](https://github.com/qualixar/superlocalmemory/issues/105); the same escape hatch embeddings got in v3.4.24, [#16](https://github.com/qualixar/superlocalmemory/issues/16)). No subprocess and no local model download. An unreachable, slow, or malformed endpoint logs an error and returns fusion-ranked results — SLM never silently substitutes the local English model. Keys, auth, and failure semantics: **[docs/configuration.md](docs/configuration.md#remote-embedding-and-rerank-endpoints)**.
+
 ### Cache + Compress
 
 <a id="three-surfaces-proxy--mcp-tools--skill"></a>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "3.8.11",
+  "version": "3.8.13",
   "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

@@ -15,5 +15,5 @@
   "mcpServers": "./.mcp.json",
   "name": "superlocalmemory",
   "repository": "https://github.com/qualixar/superlocalmemory",
-  "version": "3.8.11"
+  "version": "3.8.13"
 }

package/plugin/CLAUDE.md

@@ -1,4 +1,4 @@
-<!-- BEGIN SuperLocalMemory v3.8.11 -->
+<!-- BEGIN SuperLocalMemory v3.8.13 -->
 
 ## SuperLocalMemory (SLM) — Agent Rules
 
@@ -39,6 +39,6 @@ slm-recall · slm-remember · slm-session · slm-status · slm-cache · slm-comp
 ### Subagents
 slm-memory-advisor (memory decisions, session hygiene, scope/profile guidance) · slm-optimize-advisor (context compression + KV cache) · slm-governance-advisor (scope/roles/compliance/GDPR)
 
-<!-- END SuperLocalMemory v3.8.11 -->
+<!-- END SuperLocalMemory v3.8.13 -->
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-governance-advisor.md

@@ -77,4 +77,4 @@ slm-scope · slm-governance · slm-profile · slm-remember · slm-recall
 # What NOT to do
 Never session_init twice; never forget without dry-run preview; never store secrets; never bypass role checks; never claim an erasure succeeded without verifying via recall.
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-loop-runner.md

@@ -68,4 +68,4 @@ assessment. The gate is the authority.
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-memory-advisor.md

@@ -46,4 +46,4 @@ slm-recall · slm-remember · slm-session · slm-scope · slm-profile · slm-gov
 # What NOT to do
 Never session_init twice; never forget dry_run=False without reporting preview; never dump a whole file into remember; never invent a memory; never claim "saved" without success:true / clean CLI exit; never bypass scope or governance restrictions.
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-optimize-advisor.md

@@ -41,4 +41,4 @@ slm-compress · slm-cache · slm-status · slm-profile
 # What NOT to do
 Never compress code-for-edit/JSON-to-parse/<500 chars; never store secrets/ccr_ids; never let optimize failure block/alter the task; never claim a specific savings %; never carry ccr_ids across profile switches.
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/plugin/requirements.txt

@@ -1 +1 @@
-superlocalmemory==3.8.11
+superlocalmemory==3.8.13

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

@@ -145,4 +145,4 @@ These subcommands control daemon-level cache settings. They do not read or write
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -147,4 +147,4 @@ Content over 1 MB (1 000 000 bytes UTF-8) is processed but `reversible` is force
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -245,4 +245,4 @@ Before running any destructive operation (`forget`, `compact_memories`):
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -311,4 +311,4 @@ profile. See `slm-profile` for the full profile switching workflow.
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -96,4 +96,4 @@ paused, name the approval needed; when errored, quote the short detail.
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -279,4 +279,4 @@ mesh availability.
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -145,4 +145,4 @@ Name them differently in your MCP config (e.g. `superlocalmemory-personal` and
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -236,4 +236,4 @@ before recalling, then switch back. See `slm-profile` for workspace switching.
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -238,4 +238,4 @@ different workspace, use `switch_profile` first. See `slm-profile`.
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -173,4 +173,4 @@ to review the impact. See `slm-remember` for the full deletion discipline.
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -227,4 +227,4 @@ explicitly and call `recall` with `include_global`/`include_shared` after
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -163,4 +163,4 @@ multi-profile setup. To switch the active profile, see `slm-profile`.
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/plugin-src/rules/AGENTS.md

@@ -128,4 +128,4 @@ When the SLM MCP server is unavailable, use these CLI equivalents:
 - **slm-optimize-advisor** — context compression and KV cache
 - **slm-governance-advisor** — scope/role compliance, retention policies, GDPR
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -145,4 +145,4 @@ These subcommands control daemon-level cache settings. They do not read or write
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -147,4 +147,4 @@ Content over 1 MB (1 000 000 bytes UTF-8) is processed but `reversible` is force
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -311,4 +311,4 @@ profile. See `slm-profile` for the full profile switching workflow.
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

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

@@ -236,4 +236,4 @@ before recalling, then switch back. See `slm-profile` for workspace switching.
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -238,4 +238,4 @@ different workspace, use `switch_profile` first. See `slm-profile`.
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -227,4 +227,4 @@ explicitly and call `recall` with `include_global`/`include_shared` after
 
 ---
 
-*SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later*

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

@@ -163,4 +163,4 @@ multi-profile setup. To switch the active profile, see `slm-profile`.
 
 ---
 
-SuperLocalMemory v3.8.11 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v3.8.13 · Qualixar · AGPL-3.0-or-later

package/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "3.8.11"
+version = "3.8.13"
 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__ = "3.8.11"
+__version__ = "3.8.13"
 
 _REQUIRED_VERSIONS = {
     "sentence_transformers": "5.3.0",

package/src/superlocalmemory/cli/commands.py

@@ -22,10 +22,21 @@ logger = logging.getLogger(__name__)
 
 
 def _daemon_unavailable(command: str, use_json: bool) -> None:
-    """Exit a mutation client without opening a process-local writer."""
+    """Exit a mutation client without opening a process-local writer.
+
+    The bare "owned daemon is unavailable" of earlier releases described a
+    stopped daemon, a recycled PID, an unreachable port and an identity
+    mismatch identically, which gave issue #104's reporter nothing to act on.
+    The diagnosis names the evidence and the next command to run.
+    """
+    from superlocalmemory.cli.daemon import describe_daemon_unavailability
+
+    diagnosis = describe_daemon_unavailability()
     error = {
         "code": "DAEMON_UNAVAILABLE",
-        "message": "Owned daemon is unavailable; retry later.",
+        "reason": diagnosis["reason"],
+        "message": f"Owned daemon is unavailable: {diagnosis['message']}",
+        "hint": diagnosis["hint"],
         "retryable": True,
     }
     if use_json:
@@ -34,7 +45,8 @@ def _daemon_unavailable(command: str, use_json: bool) -> None:
         json_print(command, error=error)
     else:
         print(
-            "DAEMON_UNAVAILABLE: owned daemon is unavailable; retry later.",
+            f"DAEMON_UNAVAILABLE ({diagnosis['reason']}): "
+            f"{diagnosis['message']} {diagnosis['hint']}",
             file=sys.stderr,
         )
     raise SystemExit(1)
@@ -2282,6 +2294,28 @@ def cmd_doctor(args: Namespace) -> None:
         _check("Python", "FAIL", f"{v.major}.{v.minor}.{v.micro} (need >= 3.11)",
                "Install Python 3.11+ from https://python.org/downloads/")
 
+    # 1b. Version integrity (issue #107). Doctor is what a user runs when
+    # something seems wrong, so it is exactly where "the code you are running
+    # is not the code you installed" has to appear. Reported as WARN rather
+    # than FAIL: the installation is sound, it is this *process* that is
+    # behind, and `slm doctor` itself is short-lived so it is rarely the
+    # stale one -- it is reporting on behalf of the long-lived servers.
+    try:
+        from superlocalmemory.infra.version_integrity import check_version_integrity
+
+        _vi = check_version_integrity()
+        if _vi.is_stale:
+            _check("Version integrity", "WARN", _vi.detail, _vi.hint)
+        elif _vi.state == "mismatch":
+            _check("Version integrity", "WARN", _vi.detail, _vi.hint)
+        elif _vi.state == "unknown":
+            _check("Version integrity", "WARN", _vi.detail,
+                   "Reinstall so distribution metadata is readable.")
+        else:
+            _check("Version integrity", "PASS", _vi.detail)
+    except Exception as _vi_exc:  # noqa: BLE001 - never break doctor
+        _check("Version integrity", "WARN", f"could not verify: {_vi_exc}")
+
     # 2. Core deps
     core_modules = {
         "numpy": "numpy", "scipy": "scipy", "networkx": "networkx",
@@ -2796,6 +2830,31 @@ def cmd_mcp(_args: Namespace) -> None:
     except Exception:
         pass  # Never block MCP startup on cleanup failure
 
+    # Version integrity (issue #107). The reaper above only kills *orphans* —
+    # servers whose parent died. A server whose IDE is still alive is never an
+    # orphan, so it survives upgrades indefinitely and keeps serving the code
+    # it imported at startup. That is how eighteen `slm mcp` processes spanning
+    # four days and two releases stayed alive on one machine, and how a stale
+    # one made issue #106 look unfixed across two debugging sessions.
+    #
+    # This runs at startup, so a server launched *after* an upgrade is correct
+    # by construction and stays silent. It fires for a server that started
+    # before its own package was upgraded — which is possible when the client
+    # respawns it from an old cached path.
+    #
+    # CRITICAL: logging only, never stdout — MCP speaks JSON-RPC over stdio and
+    # any print corrupts the protocol. The logger writes to stderr.
+    try:
+        from superlocalmemory.infra.version_integrity import check_version_integrity
+
+        _vi = check_version_integrity()
+        if _vi.is_stale:
+            logger.warning("MCP server version drift: %s. %s", _vi.detail, _vi.hint)
+        elif _vi.differs:
+            logger.info("MCP server version note: %s", _vi.detail)
+    except Exception:
+        pass  # A diagnostic must never prevent the server from starting.
+
     # Auto-install hooks on MCP startup (fast path: ~0.1ms if already current)
     # CRITICAL: No stdout — MCP uses stdio transport, any print corrupts protocol
     try: