Memory in your repo
Plain markdown under .memory/ and wiki/. Git is the source of truth — the app is just a fast UI on top.
Native macOS 14+ · MCP-ready · ships in two weeks
Eight tools. One repo. Memory that never resets.
Memophant is the command center for your repo's AI memory. A basic-memory knowledge graph for atomic decisions. A wiki for long-form prose. A browsable design system. A code-symbol index with sub-second queries. A kanban for forward-looking work. Your Claude Code session history. An in-process MCP server. A runtime for agentic memory work. All in your repo, all queryable in milliseconds, all wired to Claude Code by default.
What's inside
We'll email once when it ships. That's it.
AI coding sessions forget what the last one learned. Long-form CLAUDE.md files grow stale, get truncated, and quietly rot. Memophant fixes that — by treating your repo itself as the memory store.
What it does
Memory in your repo. Health you can see. Distillation from your sessions.
Three things Memophant gets right that nothing else does, in one native app.
Health you can see
Every note remembers the commit it was written against. When the code drifts, the note shows as stale on the dashboard — memory doesn't quietly rot.
Distillation from sessions
Feed Memophant a Claude Code transcript and it proposes durable memory — merges, new relations, reclassifications. You approve; the session log dies; the lesson stays.
Layered memory
Four tiers, structured for how projects actually grow.
Notes capture atomic facts. The wiki holds long-form references. Design is the system you check before touching UI. Code is a queryable map of your symbols. Memophant keeps them separate so you write in the right place — and so every session finds what it needs without re-reading everything.
Memories — .memory/Wiki — wiki/Design — design/Tasks — TASKS.md
Session distillation
Turn a session into memory, not a transcript graveyard.
After a long Claude Code session, run Consolidate. Memophant scans the transcript, proposes merges, dedupes, new relations, and reclassifications, and shows you a before-and-after diff for every change. You apply what's right, skip what isn't.
1 · Scan
2 · Review & apply
Per-session precision guardrails: 0–3 candidates per session, mandatory citations, configurable token budget so the API spend never runs away.
Memory health
Memory you can actually see — fresh, aging, or stale.
Every note knows which files it came from and which commit it was written against. When that code drifts, Memophant flags the note on the dashboard, so the next session sees the warning before it acts on bad information.
Freshness
A 10-segment bar shows fresh / aging / stale at a glance, per project and per tier.
Drift detection
When a note's source_paths change in HEAD, the note is auto-flagged for review.
Consolidation
A one-click pass dedupes, merges superseded notes, and trims verbose instruction files.
Code intelligence
Structural code queries, in milliseconds.
Ask “where is SymbolName defined?” or “what imports this file?” and get a deterministic answer in sub-second time. Memophant indexes your code into SQLite and ships the CLI in every project — so Claude reaches for the structural query instead of grepping the whole repo.
memophant codebash
$ memophant code find TaskBoardView
Memophant/Features/Tasks/TaskBoardView.swift:14 struct TaskBoardView: View
Memophant/Features/Tasks/TaskBoardView.swift:241 extension TaskBoardView
$ memophant code refs TaskBoardView
Memophant/Features/Dashboard/DashboardView.swift:182 TaskBoardView(project: project)
Memophant/Features/Sidebar/ProjectSidebar.swift:64 NavigationLink { TaskBoardView(project: $0) }
$ memophant code status
clean · 165 files · 6810 symbols · indexed in 412msTested on Memophant itself: 165 files, 6,810 symbols, indexed in roughly 400ms. Symbol lookups complete in milliseconds — about 45× faster than the equivalent Python-toolchain query.
Run with Claude
Queue agentic memory work — and watch it run.
Audit memory drift, propose consolidations, draft wiki pages: queue them as tasks and let Claude run them in the background with the right prompt, the right tools, and a streaming log. The human reviews; the model does the legwork.
Trust
Your memory is just files in your repo.
No vendor lock-in. The memory format is plain markdown, indexed by the open-source basic-memory engine. Every commit runs a two-tier secret-scan — known token patterns block; secret-like assignments warn — so nothing dangerous reaches your wiki.
Two-tier secret-scan on every commitHeirloom License — becomes open source if Memophant is abandoned
Native macOS
Designed quiet. Built fast.
SwiftUI and SwiftData. CloudKit for the projection. Real materials, SF Pro, keyboard-driven. It opens fast, idles silently, and feels like Notes or Reminders at their best — not a cross-platform Electron shell.
Works with Claude Code
Plug Memophant into Claude Code and every session starts already up to speed.
Memophant ships an in-process MCP server. Claude Code picks it up automatically — search notes, write durable memory, query the code index — all without leaving the session. Cursor, Codex, and any MCP-aware client work the same way.
FAQ
Questions, plainly answered.
What is Memophant, in one sentence?
A native macOS app that captures what your AI coding sessions learn and keeps it in your repo as plain markdown.
Where does my memory actually live?
In your git repository, under .memory/ and wiki/. Memophant is a fast UI and orchestrator on top — your data is never trapped in the app.
Do I need Claude Code to use it?
No. The memory system works standalone via the basic-memory CLI and the included MCP server. Claude Code makes session distillation shine, but Cursor, Codex, and any MCP-aware client work too.
Is it open source?
Not today — but it's licensed under the Heirloom License, which means it's a commercial paid product while it's being actively maintained, and automatically becomes open source under MPL-2.0 if the developer ever abandons it. You buy the app; you also buy a guarantee that it won't die when its maintainer does.
What macOS versions are supported?
macOS 14 (Sonoma) and later. Apple Silicon and Intel.
How is this different from .cursorrules or a long CLAUDE.md?
Those are single files that grow stale and unsearchable. Memophant is a structured, searchable knowledge graph with drift detection, distillation, and per-tier organization — and it stays in your repo, so any client can read it.
What about my secrets?
Every commit runs a two-tier secret scan before anything reaches git. Block-on-match for known token patterns, warn-on-match for secret-like assignments. The publishable wiki gets the same gate.
When does it ship and how much does it cost?
Ships in about two weeks as a notarized DMG. Pricing announced at launch — join the list to be notified.
Ships in two weeks. Get the build the day it's notarized.
One email when it ships. No drip campaign, no upsell. Promise.