Merge pull request #110 from GRU-953/release-3.3.0

v3.3.0 — Theme-Z additive bricks (salience/confidence · provenance spans · sub-types · recall provenance)

Author: GRU-953

.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
   },
   "metadata": {
     "description": "Local, token-free, deterministic file digestion → searchable knowledge-graph memory for any AI assistant (Claude, Gemini, Cursor, VS Code, Windsurf, Codex, Grok). Incremental, portable (export/import/merge/diff), model-free (no LLM/GPU), 100% private, open-source.",
-    "version": "3.2.0",
+    "version": "3.3.0",
     "pluginRoot": "."
   },
   "plugins": [

.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "memorised-them-all",
-  "version": "3.2.0",
+  "version": "3.3.0",
   "description": "Point your AI assistant at any local files or folders: Memorised them All converts them to Markdown and digests them into a token-free, searchable knowledge-graph memory (entities, relations, themes) it can recall from — with a global synopsis, per-theme summaries, and exportable Markdown notes. Works with Claude, Gemini, Cursor, VS Code, Windsurf, ChatGPT's Codex and Grok via one `mta setup`. 100% local & private: deterministic and model-free (no LLM, Ollama, GPU, or embedding model), no network or API keys, free & open-source. Recall uses model-free BM25 ranking (script-agnostic, incl. Bengali) with a low-confidence guard; beneficiary PII is suppressed. Re-digests are incremental (only changed files are re-read); recall can filter by document or entity type and span several projects at once; and memories export, import, merge and diff across machines. A digest always finishes (per-file timeouts), memory writes are crash-safe, archives unpack safely (Zip-Slip/bomb guards), and legacy Bengali (SutonnyMJ/Bijoy) is auto-converted to Unicode with PDF-text recovery and re-OCR.",
   "author": {
     "name": "Aninda Sundar Howlader",

CHANGELOG.md

@@ -6,7 +6,16 @@ adheres to [Semantic Versioning](https://semver.org/) and
 
 ## [Unreleased]
 
-### Added (Theme-Z — accumulating toward the next major; see `program/THEME_Z_PLAN.md`)
+## [3.3.0] — 2026-06-25
+
+Theme-Z additive bricks — a richer knowledge graph + provenance, shipped ahead of the
+graph-schema-v3 major. **All additive and invariant-safe** (token-free, 100% local,
+deterministic, model-free): no breaking changes, no new tools, no graph-schema bump. New
+fields ride `graph.json` (existing consumers ignore them) and recall provenance is derived at
+query time, so the stored recall index is unchanged and **no re-digest is required** — though
+re-digesting populates the new graph fields for older memories.
+
+### Added
 - **Provenance pointers in recall (WP-134).** `recall` now surfaces the WP-123b fact spans as a
   pointer-only `spans` list (`{doc, start, end}` codepoint offsets, capped) on entity hits, so an
   assistant can cite *exactly where* a fact came from in the source `.md`. Spans are derived from

CITATION.cff

@@ -30,7 +30,7 @@ authors:
 repository-code: "https://github.com/GRU-953/memorised-them-all"
 url: "https://github.com/GRU-953/memorised-them-all"
 license: MIT
-version: 3.2.0
+version: 3.3.0
 keywords:
   - MCP
   - Model Context Protocol

manifest.json

@@ -2,7 +2,7 @@
   "manifest_version": "0.3",
   "name": "memorised-them-all",
   "display_name": "Memorised them All (local, token-free)",
-  "version": "3.2.0",
+  "version": "3.3.0",
   "description": "Convert any attachment to Markdown and digest it into a token-free, incremental knowledge-graph memory you can export, import and merge across machines — locally, deterministically, no AI models needed.",
   "long_description": "Memorised them All turns a folder of documents into durable, **token-free** memory for your AI assistant — Claude, Gemini, Cursor, VS Code, ChatGPT's Codex, Windsurf, Grok and more — entirely on your machine. It converts every attachment — PDF, Word/Excel/PowerPoint (including legacy binary .doc/.ppt/.xls via optional LibreOffice), HTML, RTF, EPUB, CSV and more — to Markdown locally using Microsoft's MarkItDown plus optional Tesseract OCR. It recursively unpacks archives (zip/tar/gz/bz2/xz natively; rar/7z when an extractor is installed) behind Zip-Slip, decompression-bomb, depth and symlink guards with content-hash dedup, and skips media/font/junk files cleanly. It then runs a fully **deterministic, model-free** pipeline — rule-based extraction of entities, relations and facts, fuzzy entity resolution, community/theme detection and layered summaries — to build a layered knowledge graph: a global synopsis, per-theme summaries, per-document Markdown notes and an exportable memory bundle. No Ollama, no LLM, no embedding model, no GPU and no network: the same corpus always produces byte-identical memory, and a digest always completes (per-file timeouts, crash-safe atomic writes). Recall uses model-free **BM25 lexical** ranking (script-agnostic, Bengali-aware) with a low-confidence guard that declines off-topic queries, and beneficiary PII (phone numbers, roster rows) is suppressed from summaries and recall. Every tool returns only compact metadata or a tiny cited slice — never document contents — so digesting and recalling whole folders costs ~0 context tokens, in whichever assistant you use.\n\nIt also converts legacy Bengali (Bijoy/SutonnyMJ ANSI fonts) to standard Unicode automatically: a font-aware OOXML delegacifier (mixed English+Bengali stays clean), line-wise PDF-text recovery, a vetted reorder-artifact repair (রম্ন→রু), and auto re-OCR (Tesseract eng+ben) of broken-font Bengali PDFs.\n\nRe-digesting is incremental: only files whose bytes changed are re-converted (a content-hash manifest tracks them) and deleted files are pruned, while the rebuilt memory stays byte-identical to a full digest. Memories are portable and composable — export to Markdown + a JSON knowledge-graph (with portable per-document content hashes) plus GraphML/CSV, diff two memories, import a bundle on another machine, or merge several projects into one.\n\nTools: digest, convert, recall, memory_overview, export_memory, list_digestible, forget, memory_status, diff_memory, import_memory, merge_memory.",
   "author": {

mta/__init__.py

@@ -3,5 +3,5 @@
 Clean-room implementation by GRU-953 (Aninda Sundar Howlader).
 """
 
-__version__ = "3.2.0"
+__version__ = "3.3.0"
 __author__ = "Aninda Sundar Howlader"

server.json

@@ -2,7 +2,7 @@
   "$schema": "https://static.modelcontextprotocol.io/schemas/2025-07-09/server.json",
   "name": "io.github.gru-953/memorised-them-all",
   "description": "Local, token-free file → knowledge-graph memory for any MCP client — Claude (Desktop & Code), Gemini CLI, Cursor, VS Code, Windsurf, ChatGPT's Codex and Grok. Converts any attachment — PDF, Office (incl. legacy .doc/.ppt/.xls), HTML, CSV, EPUB and more — to Markdown and digests it into a layered graph + exportable memory, entirely on-device. Deterministic and model-free (no LLM/embedding model/GPU), so it always completes and is byte-identical. Re-digests are incremental (only changed files are re-read); export, import, merge and diff memories across machines. 11 MCP tools; BM25 recall with document / entity-type / multi-project filters returns a tiny cited slice; PII suppressed; archives unpacked safely; legacy Bengali auto-converted to Unicode. Cross-platform (Linux/macOS/Windows).",
-  "version": "3.2.0",
+  "version": "3.3.0",
   "repository": {
     "url": "https://github.com/GRU-953/memorised-them-all",
     "source": "github"
@@ -12,7 +12,7 @@
       "registryType": "pypi",
       "registryBaseUrl": "https://pypi.org",
       "identifier": "memorised-them-all",
-      "version": "3.2.0",
+      "version": "3.3.0",
       "transport": { "type": "stdio" },
       "environmentVariables": [
         {