lean-memory 0.1.1__tar.gz → 0.1.3__tar.gz

lean_memory-0.1.3/CHANGELOG.md

@@ -0,0 +1,123 @@
+# Changelog
+
+All notable changes to lean-memory are documented here. The format is based on
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project
+adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.1.3] - 2026-07-12
+
+Publish-readiness release: an independent multi-team review of v0.1.2 found
+three launch blockers on the canonical MCP first-run path plus packaging and
+correctness majors — all fixed here.
+
+### Fixed
+- **MCP Registry install crashed on startup**: `server.json` ran
+  `uvx --from lean-memory lean-memory-mcp`, but `mcp` is an optional extra, so
+  every registry install died with `ModuleNotFoundError`. The manifest now
+  installs `lean-memory[mcp,models,extract]` (pinned by a manifest test).
+- **Model banner corrupted the MCP stdio stream**: gliner2's `from_pretrained`
+  prints a config banner to stdout — the JSON-RPC channel — on the first
+  `memory_add` of the canonical install. All model lazy-loads (GLiNER2,
+  SentenceTransformer, CrossEncoder) now route load-time chatter to stderr.
+- **Embedder swap bricked existing namespaces**: reopening a DB created with a
+  different embedder dimension (768-dim offline stub → 1024-dim Qwen after
+  installing `[models]`) failed deep in retrieval with an opaque shape error;
+  the store now refuses the mismatch at open with an actionable message.
+- **Uppercase FTS5 operator words crashed search**: `'coffee AND tea'` raised
+  `sqlite3.OperationalError` through `Memory.search` and the `memory_search`
+  tool. Terms are now quoted FTS5 string literals; the sparse arm degrades to
+  no-hits on any residual syntax error.
+- **Functional-slot supersession left stale facts current**: a replacement
+  retired only the single most-similar fact, so a slot extended by an additive
+  cue ("I also work at Globex.") kept two conflicting current employers. A
+  replacement on a functional slot now retires every co-valid latest fact;
+  multi-valued slots keep single-target retirement.
+- **High-similarity band ignored multi-valued slots**: with real embedders,
+  distinct co-valid values on a multi-valued slot (jazz/blues) embed at cosine
+  0.6–0.95 and were silently superseded; multi-valued predicates now stay
+  co-valid in every band (new resolver route: `high_extends_additive`).
+  Predicate-scoped on purpose: the textual cue ("and"/"also") stays a
+  low/mid-band signal so a conjunction-phrased replacement still supersedes.
+- **`[llm]` extra crashed every add() with Ollama stopped**: `Memory.add` now
+  catches `TyperError` and stub-types the escalated batch, as the typer
+  contract always documented.
+- **Packaging**: Apache-2.0 `LICENSE` added (repo, wheel, and sdist — the
+  license was previously declared but its text shipped nowhere); the sdist is
+  scoped to user-facing files (0.1.2 shipped internal strategy docs, agent
+  instructions, and the bench harness to PyPI); the README hero GIF uses an
+  absolute URL so the PyPI page renders it; the demo-agent flow is clone-based
+  (the script was never in the wheel).
+
+### Added
+- Schema-version stamp (`PRAGMA user_version = 1`) as the migration anchor for
+  0.1.x namespace files (pre-stamp files upgrade in place; newer stamps are
+  never downgraded).
+- `LM_FORCE_STUBS` env var pins the offline stub backends in the MCP server
+  (for tests/CI that must never load a model).
+- Subprocess-level MCP stdio protocol test: handshake + real tool call, every
+  stdout line must parse as JSON. CI matrix now covers Python 3.11/3.12.
+
+## [0.1.2] - 2026-07-12
+
+### Fixed
+- MCP Registry namespace case: `io.github.Wuesteon/lean-memory` (the registry's
+  PyPI ownership marker is compared case-exactly against the published README).
+
+## [0.1.1] - 2026-07-12
+
+### Added
+- MCP Registry metadata: `server.json` (io.github.wuesteon/lean-memory) and an
+  OIDC publish workflow; `mcp-name` ownership marker in the README (required by
+  the registry's PyPI validation — the reason for this patch release).
+
+## [0.1.0] - 2026-07-12
+
+First public release. lean-memory is an embedded, local-first agent-memory
+engine: one SQLite file per namespace, hybrid dense+sparse retrieval with
+rerank, and ADD-only supersession queryable at any past point in time
+(`as_of`). No server, no daemon, no mandatory cloud key.
+
+### Added
+
+- **MCP server** exposing memory as three tools (`memory_add`, `memory_search`,
+  `memory_clear`) for Claude Code, Claude Desktop, and other MCP clients.
+  Canonical install `pip install 'lean-memory[mcp,models,extract]'`
+  opportunistically upgrades each backend whose extra is present (real embedder
+  + reranker via `[models]`, GLiNER2 extraction via `[extract]`) and otherwise
+  falls back to deterministic offline stubs. Two-minute quickstart with
+  copy-paste Claude Code / Claude Desktop config and a demo GIF.
+- **`Memory.search(now=...)`** — recency decay now anchors to a caller-supplied
+  timestamp, so the 0.2 recency term is no longer dead on historical corpora.
+- **Point-in-time queries** via `as_of` (epoch ms) with `is_latest_only=False`.
+- **CI + release workflows** (GitHub Actions): offline test matrix on
+  ubuntu/macOS × Python 3.10/3.13, plus build-and-publish to PyPI on `v*` tag
+  via Trusted Publishing.
+- PyPI metadata: keywords, classifiers, and project URLs.
+
+### Changed
+
+- **Default embedder is now the ungated Qwen3-Embedding-0.6B** (was a gated
+  Gemma model that broke the `[models]` first run). Reranker default is
+  Ettin-32M; both are pinned ungated and covered by regression tests.
+- **Escalation engine recalibrated on real conversational turns.** Endpoint-
+  scoped coreference/ellipsis detection replaces the whole-text pronoun scan
+  (coreference escalations dropped from 65.6% to effectively nil on real
+  turns), and the `prior_entity` trigger was retired (subject re-mention is
+  normal discourse, measured at 52.8% of candidates). At the re-frozen
+  `(typing_threshold=0.4, conf_threshold=0.4)` operating point, escalation on
+  the real LongMemEval probe is **14.6%** (was ~96% pre-fix), with the residual
+  being irreducible inferential-edge (`derives`) escalations. BET-2 three-gate
+  revalidation PASSes at this operating point.
+- **Extraction granularity calibrated** — GLiNER candidate threshold set to
+  0.4, cutting the extractor from ~8 facts/turn to ~3.7 so `fact_text` reads as
+  facts rather than whole utterances.
+- MCP server loads models lazily (first tool call rather than import) so a
+  cold-cache spawn answers the MCP handshake immediately instead of blocking on
+  a model download; search output is deduplicated.
+
+### Fixed
+
+- Sparse BM25 retrieval arm now honors the `as_of` interval predicate.
+- Known-entities handed to the router/typer are capped at the 100 most recent.
+
+[0.1.0]: https://github.com/Wuesteon/lean-memory/releases/tag/v0.1.0

lean_memory-0.1.3/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

{lean_memory-0.1.1 → lean_memory-0.1.3}/PKG-INFO

@@ -1,11 +1,12 @@
 Metadata-Version: 2.4
 Name: lean-memory
-Version: 0.1.1
+Version: 0.1.3
 Summary: Embedded, local-first agent memory with hybrid retrieval and ADD-only supersession.
 Project-URL: Homepage, https://github.com/Wuesteon/lean-memory
 Project-URL: Repository, https://github.com/Wuesteon/lean-memory
 Author: lean-memory
 License: Apache-2.0
+License-File: LICENSE
 Keywords: agent-memory,embeddings,llm,local-first,mcp,memory,rag,retrieval,sqlite,vector-search
 Classifier: Development Status :: 4 - Beta
 Classifier: Intended Audience :: Developers
@@ -65,7 +66,7 @@ mem.add("user-42", "I now work at Globex.")          # supersedes Acme automatic
 mem.search("user-42", "where does the user work?")   # → "I now work at Globex."
 ```
 
-![lean-memory quickstart](docs/assets/quickstart.gif)
+![lean-memory quickstart](https://raw.githubusercontent.com/Wuesteon/lean-memory/main/docs/assets/quickstart.gif)
 
 Facts are extracted from natural language, stored in a per-namespace SQLite file, and retrieved with hybrid dense+sparse search. Old facts are never deleted — they're superseded and queryable at any past point in time.
 
@@ -111,10 +112,12 @@ mem.close()
 
 ## Demo Agent
 
-A terminal chatbot showing the full memory loop — add, retrieve, supersede, restart:
+A terminal chatbot showing the full memory loop — add, retrieve, supersede, restart.
+The demo script lives in the repo (it is not installed with the package):
 
 ```bash
-pip install 'lean-memory[examples]'
+git clone https://github.com/Wuesteon/lean-memory && cd lean-memory
+pip install -e '.[examples]'
 export ANTHROPIC_API_KEY=sk-ant-...
 python examples/chat.py                  # uses offline stubs by default
 python examples/chat.py --namespace bob  # separate memory tenant, persists across restarts
@@ -227,5 +230,5 @@ bench/                        Retrieval quality + BET-2 ablation harnesses
 
 Apache-2.0
 
-<!-- mcp-name: io.github.wuesteon/lean-memory -->
+<!-- mcp-name: io.github.Wuesteon/lean-memory -->
 

{lean_memory-0.1.1 → lean_memory-0.1.3}/README.md

@@ -21,7 +21,7 @@ mem.add("user-42", "I now work at Globex.")          # supersedes Acme automatic
 mem.search("user-42", "where does the user work?")   # → "I now work at Globex."
 ```
 
-![lean-memory quickstart](docs/assets/quickstart.gif)
+![lean-memory quickstart](https://raw.githubusercontent.com/Wuesteon/lean-memory/main/docs/assets/quickstart.gif)
 
 Facts are extracted from natural language, stored in a per-namespace SQLite file, and retrieved with hybrid dense+sparse search. Old facts are never deleted — they're superseded and queryable at any past point in time.
 
@@ -67,10 +67,12 @@ mem.close()
 
 ## Demo Agent
 
-A terminal chatbot showing the full memory loop — add, retrieve, supersede, restart:
+A terminal chatbot showing the full memory loop — add, retrieve, supersede, restart.
+The demo script lives in the repo (it is not installed with the package):
 
 ```bash
-pip install 'lean-memory[examples]'
+git clone https://github.com/Wuesteon/lean-memory && cd lean-memory
+pip install -e '.[examples]'
 export ANTHROPIC_API_KEY=sk-ant-...
 python examples/chat.py                  # uses offline stubs by default
 python examples/chat.py --namespace bob  # separate memory tenant, persists across restarts
@@ -183,5 +185,5 @@ bench/                        Retrieval quality + BET-2 ablation harnesses
 
 Apache-2.0
 
-<!-- mcp-name: io.github.wuesteon/lean-memory -->
+<!-- mcp-name: io.github.Wuesteon/lean-memory -->
 

{lean_memory-0.1.1 → lean_memory-0.1.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "lean-memory"
-version = "0.1.1"
+version = "0.1.3"
 description = "Embedded, local-first agent memory with hybrid retrieval and ADD-only supersession."
 readme = "README.md"
 requires-python = ">=3.10"
@@ -88,6 +88,35 @@ build-backend = "hatchling.build"
 [tool.hatch.build.targets.wheel]
 packages = ["src/lean_memory"]
 
+# The sdist is a public artifact: allowlist what a user or downstream packager
+# needs. Without this, hatchling ships every tracked file — which for 0.1.2
+# meant internal strategy docs (docs/superpowers/), agent instructions
+# (CLAUDE.md), and the whole bench/ harness went to PyPI.
+[tool.hatch.build.targets.sdist]
+# Leading slashes anchor to the project root (gitwildmatch): an unanchored
+# "src/" would also match e.g. .claude/worktrees/*/src/ at any depth.
+include = [
+    "/src/",
+    "/tests/",
+    "/examples/",
+    "/README.md",
+    "/ARCHITECTURE.md",
+    "/PERFORMANCE.md",
+    "/CHANGELOG.md",
+    "/LICENSE",
+    "/server.json",
+    "/pyproject.toml",
+]
+exclude = [
+    "/.claude/",
+    "/.gitignore",
+    # These test the bench/ harness (not shipped) via sys.path insertion and
+    # cannot collect without it; the fixtures are their LongMemEval-derived data.
+    "/tests/test_phase2_*.py",
+    "/tests/test_escalation_probe.py",
+    "/tests/fixtures/",
+]
+
 [tool.pytest.ini_options]
 pythonpath = ["src"]
 testpaths = ["tests"]

{lean_memory-0.1.1 → lean_memory-0.1.3}/server.json

@@ -1,19 +1,19 @@
 {
   "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
-  "name": "io.github.wuesteon/lean-memory",
+  "name": "io.github.Wuesteon/lean-memory",
   "description": "Embedded, local-first agent memory: one SQLite file per namespace, no server, no cloud.",
   "title": "lean-memory",
   "repository": {
     "url": "https://github.com/Wuesteon/lean-memory",
     "source": "github"
   },
-  "version": "0.1.1",
+  "version": "0.1.3",
   "packages": [
     {
       "registryType": "pypi",
       "registryBaseUrl": "https://pypi.org",
       "identifier": "lean-memory",
-      "version": "0.1.1",
+      "version": "0.1.3",
       "runtimeHint": "uvx",
       "transport": {
         "type": "stdio"
@@ -22,9 +22,9 @@
         {
           "type": "named",
           "name": "--from",
-          "value": "lean-memory",
+          "value": "lean-memory[mcp,models,extract]",
           "isRequired": true,
-          "description": "Install the lean-memory package, then run its lean-memory-mcp console-script entry point (the package name and script name differ, so uvx needs --from)."
+          "description": "Install lean-memory with the extras the server needs: [mcp] (imported at startup — a bare install crashes), plus [models] and [extract] (the canonical real-quality backends; ~2 GB of open models download on first tool call). The package name and script name differ, so uvx needs --from."
         }
       ],
       "packageArguments": [

{lean_memory-0.1.1 → lean_memory-0.1.3}/src/lean_memory/__init__.py

@@ -11,4 +11,4 @@ from .memory import Memory
 from .types import Entity, Episode, Fact, RetrievedFact
 
 __all__ = ["Memory", "Episode", "Entity", "Fact", "RetrievedFact"]
-__version__ = "0.1.1"  # keep in sync with pyproject.toml [project].version
+__version__ = "0.1.3"  # keep in sync with pyproject.toml [project].version

{lean_memory-0.1.1 → lean_memory-0.1.3}/src/lean_memory/embed/sentence_transformer.py

@@ -11,6 +11,9 @@ required). Both load through this one class. Requires the `models` extra
 
 from __future__ import annotations
 
+import contextlib
+import sys
+
 import numpy as np
 
 from .base import Embedder
@@ -48,7 +51,10 @@ class SentenceTransformerEmbedder(Embedder):
                     "SentenceTransformerEmbedder needs the 'models' extra: "
                     "pip install 'lean-memory[models]'"
                 ) from e
-            self._model = SentenceTransformer(self.model_name, device=self._device)
+            # stdout is the MCP stdio protocol channel — any load-time library
+            # chatter there would corrupt JSON-RPC framing on a lazy first load.
+            with contextlib.redirect_stdout(sys.stderr):
+                self._model = SentenceTransformer(self.model_name, device=self._device)
             self.dim = self._model.get_sentence_embedding_dimension() or self.dim
         return self._model
 

{lean_memory-0.1.1 → lean_memory-0.1.3}/src/lean_memory/extract/contradiction.py

@@ -98,6 +98,15 @@ _MULTIVALUED_PREDICATES = frozenset(
 )
 
 
+def is_multivalued(predicate: str) -> bool:
+    """Does this predicate naturally hold many co-valid values at once?
+
+    Multi-valued slots (likes/uses/knows/...) accumulate; everything else is
+    FUNCTIONAL — one current value, so a replacement retires the whole slot
+    (memory.py applies supersession accordingly)."""
+    return predicate in _MULTIVALUED_PREDICATES
+
+
 @runtime_checkable
 class LLMTyper(Protocol):
     """Pass-4 constrained-typing backend, narrowed to the adjudication this resolver needs.
@@ -217,6 +226,19 @@ class ContradictionResolver:
                 return Decision(
                     label=EXTENDS, target=best, similarity=best_sim, route="high_extends"
                 )
+            # Multi-valued predicates stay co-valid at EVERY band: with the real
+            # default embedder, distinct co-valid values on a multi-valued slot
+            # (jazz/blues, Python/Rust) embed at cosine 0.6-0.95, so without this
+            # check the high band silently retired facts the user still holds.
+            # PREDICATE-scoped on purpose — the textual _ADDITIVE_CUE ('and',
+            # 'also') is too weak here: at high cosine a replacement phrased with
+            # a conjunction ('I left Acme and now work at Globex.') would keep a
+            # stale value on a functional slot.
+            if is_multivalued(new_fact.predicate):
+                return Decision(
+                    label=EXTENDS, target=best, similarity=best_sim,
+                    route="high_extends_additive",
+                )
             # Same object, no added detail and not equal text ⇒ a restated change
             # (e.g. "moved to Berlin" vs "moved to Munich" that still embed close):
             # treat as a replacement on the slot.

{lean_memory-0.1.1 → lean_memory-0.1.3}/src/lean_memory/extract/gliner_extractor.py

@@ -29,7 +29,9 @@ router's escalation rate (<20% target), not the candidate count.
 
 from __future__ import annotations
 
+import contextlib
 import re
+import sys
 from abc import ABC, abstractmethod
 from typing import Any, Optional
 
@@ -161,7 +163,11 @@ class Gliner2Generator(CandidateGenerator):
             ) from exc
         # map_location="cpu" keeps us on CPU even if a checkpoint defaults elsewhere; the base
         # model is built for efficient CPU inference (no GPU required).
-        self._model = GLiNER2.from_pretrained(self.model_name, map_location="cpu")
+        # stdout is the MCP stdio protocol channel: gliner2's from_pretrained prints a
+        # model-config banner with bare print(), which would interleave with JSON-RPC
+        # framing on the first lazy load. Route load-time chatter to stderr.
+        with contextlib.redirect_stdout(sys.stderr):
+            self._model = GLiNER2.from_pretrained(self.model_name, map_location="cpu")
         return self._model
 
     def generate(self, episode: Episode) -> list[Candidate]:

{lean_memory-0.1.1 → lean_memory-0.1.3}/src/lean_memory/mcp_server.py

@@ -46,7 +46,13 @@ def _build_memory(root: Path) -> Memory:
     Each successful upgrade logs ONE line to stderr (stdout is the MCP protocol
     channel — NEVER print to stdout here) so cold-cache users, who wait through a
     ~2 GB download on first tool call, see progress in their client's server logs.
+
+    LM_FORCE_STUBS (env) pins the deterministic offline stubs even when the
+    extras are installed — for tests/CI that must never load a model.
     """
+    if os.environ.get("LM_FORCE_STUBS"):
+        return Memory(root=root)
+
     kwargs: dict = {}
 
     try: