aannoo/hcom

hcom – Hook Your Coding Agents Together

What it is

• hcom is a command‑line utility that lets you run multiple AI‑powered coding assistants (Claude, Gemini, Codex, Copilot, etc.) side‑by‑side and have them talk to each other.
• Each assistant is launched as a normal terminal process, but when prefixed with hcom the tool injects hooks that record activity in a local SQLite DB and forward messages between agents.

Why you might care

• If you frequently chain together several LLM‑based code tools (e.g., ask Claude to write a function, then have Codex review it), hcom automates the hand‑off so you don’t have to copy‑paste output manually.
• It provides a tiny dashboard (TUI) where you can watch live transcripts, file edits, and terminal screens of every agent.
• Supports spawning, forking, killing, or moving agents across machines via an MQTT relay, making it useful for distributed or multi‑device workflows.

Key features (from the README)

Feature	What it does
Real‑time messaging	Agents can send each other messages mid‑turn or when idle.
Observation	You can view status, transcripts, file changes, and live terminal output of any agent.
Event subscription	Scripts can react automatically to events such as a file edit or an agent becoming idle.
Spawn / fork / kill	Launch any number of agents in any terminal emulator (kitty, wezterm, tmux, etc.) or headless PTY.
Cross‑device relay	Connect agents on different machines through an MQTT relay using a token.
CLI & TUI	hcom command provides many sub‑commands (send, list, events, run, …) and a dashboard you can open with just hcom.
Configurable	Settings live in ~/.hcom/config.toml; you can set defaults, per‑agent overrides, and even isolate state per project.
Workflow scripts	Bundled scripts (debate, confess, fatcow, onidle) and a simple plug‑in system for custom multi‑agent patterns.

Typical workflow

1. Open two terminals.

   hcom claude   # launches Claude in a terminal window
   hcom codex    # launches Codex in another window
   

2. In either window, type a prompt like “ask the other agent their favorite cake”.
3. The agents can now exchange messages, spawn sub‑agents, or wait for each other to become idle, all orchestrated by hcom.
4. Open the dashboard with hcom to monitor everything in one place.

Installation

• Via the Python‑centric uv tool: uv tool install hcom (cross‑platform).
• Homebrew formula: brew install aannoo/hcom/hcom.
• Direct installer scripts for macOS/Linux (curl … | sh) or Windows PowerShell.

Supported AI tools (automatic integration)

• Claude, Gemini, Codex, OpenCode, Kilo, Pi, Oh My Pi, Cursor, Kimi, Copilot, Qoder, Grok, Antigravity, plus any other CLI that can be wrapped with hcom start.

When it’s useful

• Building pipelines where one LLM generates code and another reviews or tests it.
• Running long‑running background agents that need to be nudged when idle.
• Coordinating many small AI assistants across several terminals or remote machines.

License

• MIT, so you can freely modify and redistribute.

---

All details above are taken directly from the repository’s README.