Tools/silverstein/minutes

minutes

Every meeting, every idea, every voice note, searchable by your AI. Open-source, privacy-first conversation memory layer.

1.6k ★+13/wkemergingRustMIT License →trending

View on GitHub →Compare in AI Meeting Assistants →Homepage →

Replaces:Otter.ai

See how it compares →

The Lens

By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.

Updated Oct 2026

Minutes turns your meetings, calls and voice memos into Markdown files your AI assistant can search. It records on your machine, transcribes locally, can tag speakers, and writes each conversation to ~/meetings/ with decisions and action items in the frontmatter. Then Claude Code, Codex, Cursor or any MCP client can tell you what you decided last month. MIT licensed Rust, and the author promises no paid tier.

There's a signed desktop app for Apple Silicon Macs and Windows, plus Homebrew, cargo install and a prebuilt Linux CLI. The Mac app transcribes with Parakeet; other installs use Whisper. Summaries are off by default: your agent summarizes over MCP, or you point it at Ollama, a Claude or Codex subscription, or an API key.

Solo: free, best if you live in Claude Code or Obsidian. Small teams: sharing means sharing a folder. Larger teams on Granola Business at $14 per seat get shared workspaces and admin controls Minutes doesn't attempt.

The catch: with no bot, nobody else on the call is told automatically, and in all-party-consent states like California getting permission is on you, though Minutes reminds you by default. It's also pre-1.0 and ships releases almost daily, so expect churn.

Free vs Self-Hosted vs Paid

fully free

Free: MIT licensed with no paid tier; the README says "no relicensing, no paid tier for anything in this repo." You get the desktop app (menu bar capture, dictation, Recall, documents, Coach), a CLI, an MCP server (npx minutes-mcp), a Claude Code plugin and a TypeScript SDK.

Self-hosted: Everything runs on your machine; there is no server. Signed desktop app for Apple Silicon Macs (Parakeet v3 via a bundled sherpa plugin) and Windows (zip, needs the Visual C++ runtime). Homebrew CLI, cargo install minutes-cli and the prebuilt Linux binary (glibc 2.39+) default to Whisper: small is 466MB (recommended), base 141MB, tiny 75MB. Optional speaker diarization adds about 34MB of ONNX models (pyannote-rs).

Paid: No paid tier. Summarization defaults to none (your agent summarizes over MCP). Optional engines: a Claude Code, Codex, OpenCode or Pi subscription, local Ollama, any OpenAI-compatible endpoint, or a direct Claude/OpenAI API key, each at that provider's normal rates.

Free with no paid tier. The only cost is a cloud model if you pick one for summaries; Ollama or the agent subscription you already pay for keeps it at zero extra.

What to do by team size

Solo: free; desktop app or Homebrew, five minutes
Small team: free; share the Markdown folder
Medium team: free per person, but no shared workspace or admin; Granola Business at $14/user/mo if you need one
Large team: free per person, but no SSO or retention controls; look at Granola Enterprise at $35/user/mo or anarlog Enterprise

Self-hosting ops:trivial

Get tools like this every Wednesday

One featured tool, three on the radar. No fluff.