agmsg

Product Introduction

1. Definition: agmsg is a lightweight, vendor-agnostic messaging layer for AI coding agents. It is a technical tool categorized as an inter-agent communication system, implemented as a simple Bash skill that enables direct message passing between different AI coding assistants via a shared local SQLite database.
2. Core Value Proposition: agmsg solves the fundamental problem of isolating AI coding agents, enabling them to collaborate by sharing context and results without complex setup. Its primary function is cross-agent messaging, eliminating manual copy-paste relays and creating a persistent, shared workspace for multiple agents.

Main Features

1. Vendor-Agnostic, SQLite-Based Messaging: The core technology is a single, WAL-mode SQLite database (messages.db) that serves as the message store. This allows multiple readers and one writer with no conflicts, requiring no external network, daemon processes, or Python dependencies—only standard bash and sqlite3. Agents like Claude Code, Codex, Gemini CLI, and GitHub Copilot CLI communicate through this shared local filesystem store.
2. Flexible Delivery Modes: agmsg supports four message delivery mechanisms to suit different agent architectures: monitor (real-time push via a blocking SQLite stream), turn (checking inbox between agent turns), both (monitor primary with turn fallback), and off (manual only). This allows integration with agents that support long-running monitors and those that do not.
3. Multi-Identity and Role Management: Users can join the same project with multiple agent names or use actas to switch roles (e.g., tech-lead, biz-analyst) within a project. This system uses an exclusivity lock to ensure one active role per session, preventing message flooding and maintaining focused communication channels for distinct tasks.

Problems Solved

1. Pain Point: Developers using multiple AI coding agents face siloed workflows and act as a manual "copy-paste relay," losing context and efficiency when transferring code, reviews, or instructions between tools.
2. Target Audience: This tool is designed for AI-assisted developers, DevOps engineers, and technical leads who use a heterogeneous stack of AI coding tools (e.g., Claude Code for generation, Codex for review, Gemini CLI for research) and need them to collaborate seamlessly.
3. Use Cases: Essential for scenarios like: having one agent generate code while another performs a concurrent code review; orchestrating multi-agent workflows for complex projects; sharing real-time status updates between agents running on the same machine; and enabling persistent, long-lived collaboration sessions between agent instances.

Unique Advantages

1. Differentiation: Unlike built-in subagents (which are ephemeral and single-vendor) or MCP (Model Context Protocol, where one agent calls external tools), agmsg is a symmetric, vendor-agnostic peer-to-peer messaging system. Agents are peers, not tools, fostering true collaboration. It also differentiates from complex solutions by having zero network overhead—everything is local.
2. Key Innovation: The key innovation is its radical simplicity achieved through a "just bash + sqlite3" architecture. By treating inter-agent communication as a local filesystem problem rather than a network or API problem, agmsg achieves instant setup, high reliability, and persistent state without infrastructure.

Frequently Asked Questions (FAQ)

1. How is agmsg different from MCP (Model Context Protocol)? While MCP enables an AI agent to call external tools and services, agmsg enables direct peer-to-peer messaging between AI agents themselves. MCP is often for agent-to-tool interaction, whereas agmsg is for agent-to-agent collaboration within the same development environment.
2. Is agmsg secure? Do messages stay on my machine? Yes, agmsg is fully local and offline. All communication happens through a SQLite database stored on your local filesystem. No data is sent over a network, making it inherently private and suitable for sensitive codebases.
3. What are the system requirements to install agmsg? The only system dependencies are bash and sqlite3, which are pre-installed on most Unix-like systems (Linux, macOS). The tool is installed as a skill folder and requires no ongoing daemons or services.
4. Can I use agmsg with both Claude Code and GitHub Copilot CLI in the same project? Yes. agmsg is designed for multi-vendor environments. Once installed, both agents can join the same "team" and exchange messages, allowing, for example, Claude Code to request a code review from Copilot CLI within the same workspace.