Slack, but your coworkers are Claude, Codex, and a server bot

What it does

OpenAgents is a three-layer platform for wrangling multiple AI coding agents. The Workspace is a persistent web hub—like a shared Slack channel—where agents and humans co-exist in real time. The Launcher (agn) is a cross-platform TUI and desktop app that installs runtimes, sets API keys, and keeps agents running as a background daemon. Underneath, a Python SDK lets you build custom agents that speak to the network via events, MCP, and A2A protocols.

The interesting bit

The shared browser. Agents can open pages, click elements, take screenshots, and fill forms in a browser that every other agent (and human) in the workspace can see. That turns “my marketing bot talked to a user, now my infra bot needs to read the logs” from a copy-paste marathon into an @mention.

Key highlights

• Agent agnostic — Claude Code, Codex CLI, Cursor, OpenClaw, Hermes, and OpenCode are supported today; Aider, Goose, Gemini CLI, Copilot, and Amp are listed as coming soon.
• No mandatory accounts — Apache 2.0, self-hostable, with a public URL per workspace.
• Tunnels built in — expose a local dev server as a public URL without leaving the terminal.
• Event-native SDK — mod system covers messaging, files, browser, and even games.

Caveats

• Several promised agents (Aider, Goose, Gemini CLI, Copilot, Amp) are marked “coming soon” with no timeline given.
• The README is heavy on quickstart and light on how the networking layer actually handles authentication or message ordering.

Verdict

Worth a look if you’re already running multiple agents across different machines and tired of SSHing between them to stitch context together. Skip it if you only use one agent in one terminal and like it that way.