agmsg

What is agmsg?

agmsg is a cross-agent messaging tool for CLI-based AI agents. It lets agents such as Claude Code, Codex, Gemini CLI, and GitHub Copilot CLI exchange messages through a shared local SQLite database instead of relying on a separate server, broker, or human copy-paste workflow.

The project is designed for peer-to-peer coordination across different agent sessions in the same team or workspace. It supports persistent message history, team-based rooms, and role switching so agents can communicate, resume old conversations, and share context across sessions.

Key Features

• Shared local SQLite transport: messages are appended to a local database, which keeps communication offline and avoids a separate daemon or network service.
• Cross-agent messaging: CLI agents can send and receive messages across different tools, including Claude Code, Codex, Gemini CLI, and GitHub Copilot CLI.
• Delivery modes tied to agent capabilities: the project supports hook-based delivery and monitor-mode streaming, with defaults that vary by agent.
• Persistent history and replay: messages remain available after a session ends, and history.sh can replay a room into a fresh agent session.
• Team and agent setup on first use: the command prompts for a team name and agent name, then registers the project for that identity.
• Role management with actas and drop: users can switch between named roles within the same project and release locks when roles are no longer needed.

How to Use agmsg

Install it with the provided setup or install script, then restart your CLI agent so it picks up the new skill or command. On first run, launch the command in your agent, enter a team name and agent name, and choose the delivery mode if prompted.

After setup, you can send messages naturally in the agent interface, such as asking it to message another agent, check incoming messages, or list who is on the team. For automation and advanced workflows, the repository also exposes script-level commands and project/team management utilities.

Use Cases

• Coordinating code review between two CLI agents that are working in the same repository but in different sessions.
• Passing task updates between agents without copy-pasting prompts or results through a human operator.
• Keeping a durable project conversation log that can be resumed in a new session after the original agent exits.
• Running a multi-role workflow in one project, such as using one identity for architecture review and another for requirements analysis.
• Replaying prior team history into a fresh agent to restore context for follow-up work or handoff.

FAQ

Does agmsg require a server? No. The project description says it uses a shared local SQLite database and does not require a daemon, socket broker, or network service.

Is agmsg the same as MCP or subagents? No. The source explicitly says it is not MCP and not a subagent system; it connects peer sessions across tools instead of spawning child processes.

What happens on first use? The command prompts for a team name and an agent name, then asks you to pick a delivery mode based on the agent and setup.

Can message history be kept after a session ends? Yes. The source says history is durable and can be replayed into a new agent session.

Alternatives

• Manual copy-paste between agent sessions: simplest fallback, but it requires a human to relay every message and does not preserve a shared local history.
• A traditional message queue or broker: better suited to infrastructure-heavy workflows, but it is a different architecture from agmsg’s file-backed local SQLite approach.
• MCP-based integrations: useful when you want a model context protocol server, but agmsg explicitly positions itself outside that pattern.
• Subagent or child-process workflows: appropriate for spawning helper agents within one system, whereas agmsg is built for peer sessions across separate CLI tools.