silverstein/minutes

What it solves

Minutes provides a private, locally-owned conversation memory layer. It solves the problem of fragmented meeting notes and the lack of a secure, queryable archive of human decisions, intent, and context that AI agents can access without compromising privacy or relying on cloud-based notetakers.

How it works

Minutes captures audio from microphones or system audio (on macOS), transcribes it locally using whisper.cpp, performs speaker diarization via pyannote-rs, and optionally summarizes the text using an LLM (local or cloud). The resulting data is stored as plain markdown files with structured YAML frontmatter in a local directory (~/meetings/). These files are then exposed to AI agents via a Model Context Protocol (MCP) server, a CLI, or a dedicated desktop app, allowing agents to query the user's conversation history.

Who it’s for

• Agent-first users: People who use AI agents like Claude Code, Cursor, or Codex and want their agents to have context from their real-world meetings and voice memos.
• Local-first users: Individuals who重視 privacy and prefer their data to remain on their own disk, ensuring their memory archive is searchable via grep and future-proof

Highlights

• Local-first privacy: Audio is transcribed and processed locally; text only leaves the device when explicitly authorized for cloud summarization or agent context.
• MCP Integration: Works as an MCP server, making conversation memory available to any MCP-compatible AI client.
• Structured Output: Saves meetings as markdown files with YAML frontmatter for action items, decisions, and participants.
• Cross-meeting intelligence: Ability to research topics, track commitments, and others across multiple meetings.
• Multi-surface access: Available as a macOS desktop app, CLI, and MCP server.
• Voice memo pipeline: Automatically processes audio files synced from a phone to a desktop via cloud storage (e.g., iCloud).
• Consent disclosure aid: Built-in tools to help users manage and document participant consent for recording.