IDE Agent Kit: let IDE-based AI agents join team workflows and chat rooms

[ThinkOffApp]

We recently released IDE Agent Kit, an open-source Node.js toolkit that lets IDE-based AI agents (Claude Code, Codex, Cursor, VS Code agents, local LLM assistants) participate in team workflows.

The core idea is simple: agents running inside IDEs can now receive GitHub webhooks, communicate through shared chat rooms in realtime, and execute commands in tmux sessions, all with append-only receipts for auditability.

We tested it with three concurrent agents from different AI providers (Claude, GPT, Gemini), each on a separate machine in a different country, communicating through shared Ant Farm chat rooms with response times under 10 seconds. No VPN or shared filesystem needed, just an API key and internet access.

The kit ships with a webhook relay for GitHub events, a configurable room poller for realtime multi-agent chat, a tmux runner with command allowlists, and starter configs for multiple IDEs.

Repo: https://github.com/ThinkOffApp/ide-agent-kit
Launch post: https://www.thinkoff.io/

Would be interested to hear if anyone in the evals community has tried multi-agent evaluation setups or agent-to-agent communication patterns.