Minutes

Minutes is an open-source, self-hosted web app and server that captures meetings and voice memos. It functions as a local conversation memory system that transcribes audio and converts it into structured markdown files. The software is available as a macOS desktop application, a command-line interface, and an MCP server, allowing users to choose the interface that best fits their workflow.

Users can record live audio from microphones or system audio on macOS, process existing audio files, and take timestamped notes during conversations. The system stores all data as plain markdown files in a local directory, ensuring that the conversation history remains accessible via standard tools like grep or markdown-based knowledge bases. This approach removes reliance on proprietary SDKs or API keys for data retrieval.

Key features

• Local transcription using whisper.cpp and parakeet
• Native audio capture for Zoom, Meet, and Teams on macOS 15+
• Speaker identification through native diarization
• 31 MCP tools for AI agents to query conversation history
• Relationship indexing to track commitments and people across meetings
• Support for sensitive meetings with typed markers instead of audio
• Consent disclosure aids to track participant notification
• Full-text search and filtering for action items and commitments

The software is built with Rust and utilizes a pipeline that moves from audio capture to transcription, diarization, and LLM-based summarization. It integrates with various AI agents and clients, including Claude Desktop, Cursor, and Gemini CLI, through the Model Context Protocol (MCP). It also supports integration with Obsidian and Logseq by writing files directly to the local disk. For those using the CLI, the software provides commands to manage recording devices, apply summary templates, and rebuild relationship indexes.

Minutes is local infrastructure for AI agents, prioritizing data ownership and privacy by keeping audio and transcripts on the user's machine.