What is agmsg?

agmsg is a cross-agent messaging tool that lets CLI-based AI coding agents like Claude Code, Codex, Gemini CLI, and Copilot CLI communicate directly with each other. Instead of acting as a human copy-paste relay between agents, agmsg uses a shared SQLite database as the backbone — no daemon, no network service, and no Python required. It installs as an Agent Skill with just bash and sqlite3, making it lightweight and dependency-free.

Who it's for

• Solo developers running multiple agents — coordinate Claude Code and Codex in the same project without manual handoffs.
• Teams experimenting with multi-agent workflows — let different AI agents take on distinct roles (e.g., coder, reviewer, architect) and message each other autonomously.
• Advanced users building agent-to-agent pipelines — set up persistent, vendor-agnostic communication between agents that survives session restarts and project switches.

Key features

Vendor-agnostic messaging

Unlike built-in subagents that are tied to a single vendor and ephemeral by nature, agmsg works across Claude Code, Codex, Gemini CLI, and Copilot CLI. Agents join a shared "room" identified by team and agent name, and messages persist in a SQLite database — no lock-in, no disappearing context.

No daemon, no network, no Python

The entire system runs on bash and sqlite3, installed as a skill in under a minute. There is no background process to manage, no network port to configure, and no Python runtime to install. This makes it trivial to set up and tear down across different machines and projects.

Multiple delivery modes

Agents can receive messages in real-time via a monitor mode (push-based) or check between turns (poll-based). Claude Code defaults to monitor mode for instant delivery, while Codex uses turn-based checking since it lacks a Monitor tool. You choose the mode that fits each agent's capabilities.

Role-based identity switching

A single project can host multiple agent identities with different roles — for example, a tech-lead identity for architecture reviews and a biz-analyst identity for requirements work. Switching between roles is done with a simple command, and each role maintains its own message history.

What stands out

agmsg lets your AI agents talk to each other so you don't have to be the messenger.

This captures the core value: instead of manually copying output from one agent and pasting it into another, agmsg automates the handoff. Two Claude Code instances left alone in the same team can play tic-tac-toe against each other with no human in the loop — each picks up the other's move in real time. In production use, Claude Code can ask Codex for a code review and get the results back automatically, all over agmsg.

Worth checking out if…

You're tired of being the human relay between your AI coding agents, or you want to experiment with multi-agent workflows without committing to a single vendor's ecosystem. agmsg is especially useful if you value simplicity — no daemon, no network, no Python — and want a solution that works out of the box with bash and sqlite3. If you run multiple CLI agents in the same project and wish they could coordinate autonomously, agmsg is worth a look.