At a Glance:
Minutes is an open-source conversation memory tool that captures meeting audio, transcripts, decisions, and commitments locally as markdown files, making human context queryable for AI agents through CLI, MCP server, and desktop app surfaces.
Overview:
Minutes is a local-first conversation memory infrastructure designed to capture the human side of agent workflows—meetings, voice memos, intent, and decisions—and store them as plain markdown files in ~/meetings/. It provides audio recording with speaker diarization, transcript generation, and a relationship intelligence system that tracks people, commitments, and topics across all conversations. Rather than functioning solely as a meeting notes app, Minutes positions itself as queryable memory for AI coding agents (Claude Code, Codex, Gemini CLI, OpenCode, Pi) through MCP tools, CLI commands, live transcript streaming, and direct file access without requiring API keys or SDK integration.
Key Decision Points:
Local-first by default: Audio never leaves the machine unless cloud LLM summarization is explicitly enabled, and all meeting data is stored as plain markdown files.
Multiple interaction surfaces: Users can choose between a desktop app for recording and live capture, an MCP server for agent-driven search and recall, a CLI for terminal workflows, and agent-specific plugins for Claude Code and OpenCode.
Plain file storage with no vendor dependency: Meeting artifacts are written to
~/meetings/as markdown, accessible to any AI tool that reads files directly without requiring an SDK or API key.Relationship graph indexing: A SQLite index tracks people, commitments, topics, and relationship health across meetings, supporting queries about past promises and detecting lapsed contacts.
macOS system audio limitation requires workaround: Recording both sides of a call on macOS requires installing BlackHole and configuring a Multi-Output Device since system audio capture is restricted.
Core Features:
Meeting recording: Captures audio from calls (Zoom, Meet, Teams, Webex) with speaker identification via native diarization, with a BlackHole-based workaround for macOS system audio.
Voice memo processing: Ingests voice recordings from phone inbox folders through a file watcher service, supporting iOS Shortcuts, Android cloud sync, Syncthing, and manual transfer methods.
Relationship intelligence: Tracks people, commitments, topics, and relationship health using a SQLite index rebuilt from markdown, detects duplicate contacts, and surfaces alerts when losing touch with someone.
Search and query: Supports natural-language queries about past conversations, commitments, and people through MCP tools and CLI commands.
Obsidian/Logseq vault sync: Provides symlink, copy, and direct-write strategies for integrating meeting data into personal knowledge bases, with automatic vault detection.
Agent-native integration surfaces: Exposes MCP tools for meeting search, relationship memory, live transcript streaming, consistency reporting, and voice profiles, plus agent-specific plugins and skill packs.
Use Cases:
Developers using AI coding agents who want to query past meeting commitments and decisions directly from Claude Code, Codex, OpenCode, or Gemini CLI through MCP tools or file-based access.
Individual knowledge workers who record meetings and voice memos and want locally stored, privacy-preserving transcripts with relationship tracking across conversations.
Terminal-first users who prefer CLI-based meeting processing, search, and vault synchronization over GUI tools.
Obsidian/Logseq users who want to integrate meeting notes and relationship intelligence into their existing personal knowledge base through file-based sync strategies.
Open-Source Alternative Value:
Minutes provides an open-source alternative to cloud meeting memory tools by storing all conversation data locally as plain markdown files rather than in proprietary cloud services. Its architecture exposes meeting artifacts, transcripts, and relationship graphs through file-based access, MCP tools, and CLI commands without requiring SDK integration or API keys, allowing any AI agent that reads markdown files to access conversation history. The project uses the MIT license and includes integration surfaces for Claude Code, Codex, OpenCode, Gemini CLI, and other MCP-compatible agents.
