Memophant v1.1.0
Memophant 1.1
Memophant 1.1 is about trustworthy memory: your project's memory now knows when it's stale, heals itself deterministically, and converges toward a consistent, searchable shape — while every AI-proposed change stays behind your review.
Memory that knows when it's stale
- Freshness is grounded in code drift, not the calendar. Notes anchor to the files they document (
source_paths); when that code changes, the note is flagged — a note that hasn't changed because nothing it describes changed stays green forever. - Grounding happens for you. Agents declare their sources at write time, Memophant infers anchors from the files a note names, and Consolidate grounds unanchored notes automatically. One-click Accept / Choose Files… / Mark Durable actions on every note.
- Honest signals. Retired notes stop nagging, orphaned notes (whose source code was deleted) get their own signal, timed-out audit batches are never green-washed, and one-click Backfill stamps created/updated dates from your git history.
A consolidation engine you can trust
- Scope-bounded agentic audits run in small batches with per-batch timeouts, investigate only evidence-backed drift, and resume exactly where an interrupted pass left off.
- An adversarial verifier independently re-checks every proposed change against your actual code before it reaches your review — refute-by-default, with the reason shown.
- The structure lane converges drifted prose-only notes into the hybrid shape (prose kept verbatim + distilled
## Observations+## Relations) a few notes per pass, with honest reporting of what was examined. - Everything runs on your Claude subscription — consolidation, doc generation, migration, and task runs use the Claude CLI; no separate API key, no metered spend.
Memory, wiki, and code stay current together
- Memory → wiki history promotion: significant transitions (tech swaps, reversed decisions, re-architectures) move into a curated
## Historyon the topic wiki page, with bidirectional links — and when the source note drifts past the captured snapshot, re-promotion is proposed automatically. - Code tier goes multi-language: TypeScript, JavaScript, Python, Go, and Rust symbol extraction via tree-sitter, with per-page drift dots in the Code browser.
- Per-page provenance and freshness dots across wiki and code pages, with drift baselines tied to your review acknowledgments.
A stronger agent surface (18 → 27 MCP tools)
- New tools:
memory_health,review_memory,retire_memory,heal_memory,get_task,list_vendors, documents/templates tier access, and code-index staleness signals. - Writes are fast and safe: single-note indexing eliminated the full-corpus re-embed hangs; titles with slashes or punctuation can no longer create phantom folders or invisible files; a structural nudge (never an error) coaches agents toward the memory grammar with ready-to-run fix calls.
- The server ships prebuilt inside the app — no Swift toolchain needed — self-reports its build identity, and tells agents when a rebuilt binary needs a session restart.
Quality of life
- Notes render their full body — long-form memos with custom sections are no longer hidden — and display their real titles.
- Semantic search now reads prose notes, not just structured bullets.
- Project-wide task board works from any git worktree, with priority filter/sort and safer concurrent edits.
- Sidebar width persists across launches; clearer health bar; the download is a notarized, stapled DMG.
Full commit history: v1.0.1…v1.1.0