About this project

hcom is a command-line tool that connects AI coding agents running in different terminals, allowing them to communicate, monitor each other, and spawn or manage subagents. It works by wrapping agent launches (e.g., `hcom claude`, `hcom codex`) and recording activity to a local SQLite database, which agents can query to send messages, observe status and transcripts, and react to events. Key features include: - Real-time messaging between agents, with wake-on-idle support. - Observation of agent status, transcripts, file edits, live terminal screens, and command history. - Event subscriptions for status changes, file edits, collisions, and custom triggers. - Spawning, forking, resuming, and killing agents in various terminal emulators or headless mode. - Cross-device coordination via an MQTT relay with end-to-end encryption (XChaCha20-Poly1305) and a shared pre-shared key. - A TUI dashboard (`hcom`) for managing agents. - Bundled workflow scripts (e.g., debate, confess, fatcow, onidle) and support for user-defined scripts. - Configuration via `~/.hcom/config.toml` with per-agent overrides and environment variable support. Supported AI tools include Claude Code, Gemini CLI, Codex CLI, Antigravity CLI, OpenCode, Kilo Code, Pi, Oh My Pi, Cursor CLI, Kimi, Copilot CLI, and Grok Build. Other tools can join via `hcom start` and `hcom listen`. The project is written in Rust and licensed under MIT. Installation options include `uv tool install hcom`, Homebrew, and shell/PowerShell installers.