Your AI remembers every
conversation you've had

Record meetings. Capture voice memos. Ask Claude what was decided three weeks ago. Everything runs locally. Open source, free forever.

Terminal
$ minutes record --context "Q2 pricing with Alex"
Recording... (Ctrl-C or `minutes stop` to finish)
Tip: add notes with `minutes note "your note"`
$ minutes stop
Transcribing... Saving...
Saved: ~/meetings/2026-03-17-q2-pricing-with-alex.md
$ minutes search "pricing"
2026-03-17 — Q2 Pricing with Alex [meeting]
[4:20] I think monthly billing makes more sense...
Desktop app brew install --cask silverstein/tap/minutes
Copied!
CLI only brew tap silverstein/tap && brew install minutes
Copied!
MCP server (no Rust needed) npx minutes-mcp
Copied!

macOS, Windows, Linux. npx works everywhere — Claude Desktop, Cursor, Windsurf, any MCP client.

How it works

Audio  →  Transcribe  →  Summarize  →  Structured Markdown
          (local)        (your LLM)     (decisions, action items,
         whisper.cpp    Claude / Ollama   people, entities)

Your audio never leaves your machine. Transcription is local via whisper.cpp. Summarization is optional — Claude does it conversationally when you ask, using your existing subscription. No API keys needed.

Built for everyone who has conversations

Developers

15 CLI commands. 124 tests. Rust engine, single binary, MIT license. Homebrew, cross-platform CI, cargo install. TypeScript SDK for agent developers.

Knowledge workers

Menu bar app with one-click recording. Calendar integration suggests recording before meetings. Voice memo pipeline from iPhone. Obsidian vault sync.

AI agents

13 MCP tools. 6 resources. Structured intents in YAML. Decision consistency tracking. People profiles. Any agent that speaks MCP can use Minutes as its memory layer.

What you get

>
Local transcription — whisper.cpp on Apple Silicon. Your audio stays on your machine.
>
Speaker diarization — pyannote separates "who said what" in multi-person meetings.
>
Structured extraction — action items, decisions, and commitments as queryable YAML, not buried in prose.
>
Cross-meeting intelligence — minutes research "pricing" searches all meetings. minutes person "Alex" builds a profile.
>
Decision consistency — flags when Tuesday's decision contradicts Thursday's. Catches drift humans miss.
>
Voice memo pipeline — iPhone Voice Memos → iCloud → auto-transcribe on Mac. Ideas while walking the dog, searchable by afternoon.
>
Claude-native — MCP server for Claude Desktop, Cowork, Dispatch. Claude Code plugin with 11 skills. No API keys — your existing Claude subscription.
>
Any LLM — Ollama for local. OpenAI if you prefer. Or skip summarization entirely — the transcript is the artifact.
>
Markdown is the truth — every meeting saves as markdown with YAML frontmatter. Works with Obsidian, grep, QMD, or anything.

How it compares

Granola Otter.ai Meetily minutes
Local transcription No No Yes Yes
Open source No No Yes MIT
Free $18/mo Freemium Free Free
AI agent integration No No No 13 MCP tools
Cross-meeting intelligence No No No Yes
Voice memos No No No iPhone pipeline
People memory No No No Yes
Data ownership Their servers Their servers Local Local