agmsg
Cross-agent messaging for CLI AI agents, using a shared local SQLite database with no daemon and no network.
What is agmsg?
agmsg is an open-source tool that lets command-line AI agents such as Claude Code, Codex, Gemini CLI and GitHub Copilot CLI exchange messages with each other. Communication happens through a shared SQLite database on the local filesystem, so there is no daemon, network service or external infrastructure to run. It supports team-based identities, multiple roles per project and several delivery modes that suit different agent capabilities.
Key Features
Shared SQLite messaging
Agents exchange messages through a local SQLite database running in WAL mode, with no daemon or network required.
Multi-agent support
Works across Claude Code, Codex, Gemini CLI, GitHub Copilot CLI and Antigravity.
Delivery modes
Four modes (monitor, turn, both, off) let each agent choose between real-time push, between-turn checks, a fallback combination or manual checking.
Team-based identities
Agents are organised into teams with a viewable roster and message history.
Role switching
The actas command switches roles within a project and drop removes a role, with session-scoped exclusivity locks.
Cross-project identity reuse
Identities can be reused across different projects.
One-line install
A single curl-piped setup script installs the tool, with an alternative git clone and install.sh path.
Pros & Cons
Advantages
- It is fully open source under the MIT licence, so it can be inspected, modified and self-hosted freely.
- There is no network service or daemon to maintain, which keeps the setup and attack surface small.
- It works across several different CLI AI agents rather than being tied to one vendor.
- The delivery modes give flexibility for agents that do or do not support real-time monitoring.
- Installation is quick via a single one-line command.
Limitations
- It depends on bash and sqlite3 and is built almost entirely in shell, which limits it largely to Unix-like environments.
- Because messaging runs through a local SQLite database, agents must share the same filesystem rather than communicating across machines over a network.
- The project provides no commercial support, hosted service or formal documentation site beyond the GitHub README.
- Real-time monitor mode carries around five seconds of latency and relies on agent-specific hooks.
Use Cases
Developers running several CLI AI agents who want them to coordinate on a shared task without manual copy-pasting between terminals.
Teams building autonomous multi-agent workflows that need agents to hand off work and exchange context locally.
Engineers who want inter-agent messaging without standing up any server, queue or network infrastructure.
Users mixing different agents such as Claude Code and Codex who need a common channel that all of them can read and write.
Hobbyists and tinkerers experimenting with multi-agent setups on a single machine.