Introducing Agent Radio - a simple local service that makes it easy for different AI agent sessions to speak to eachother and coordinate. While working on projects, I found myself building exec plans and copy-pasting between terminals for different repositories and agents often, as well as different harnesses that were better for different things. Agent Radio is my lightweight solution to that.
No reviews yetBe the first to leave a review for Agent Radio
Maker
📌
Sub-agents are great, but are generally a feature of the CLI/harness - quick sandboxes to minimise context bloat. What I wanted was for my backend agent in repo A to be able to talk with a frontend app agent in repo B, then orchestrate changes on both sides, run tests and give feedback to eachother.
I used H5i briefly and it inspired me to build out a lightweight agent bus for my purposes.
Agent Radio lets any agent message another, allowing for simple coordination of work, and less time manually passing messages from one place to another. It's for developers who have components split across repositories, where direct coordination unblocks momentum.
Another thing I wanted was to manage and spawn set agents/harnesses in different repositories. So it comes with a tmux session control panel that lets you also template workspaces and jump in and out of them from one TUI.
Should be compatible with any harness/CLI, since it relies on tmux and direct terminal injection.
Let me know if you have any questions, discover any bugs or want to contribute to it. Open source so do as you please, have fun building 🚀
Report
Curious how this actually handles sync when two sessions are mid-execution at the same time — is it message queue based or just polling? Also wondering if it stays entirely local or if any coordination data ever leaves my machine.
Report
love how lightweight this stays, no bloated dashboard or extra ceremony, just the agents talking to each other cleanly. the fact that it came out of real workflow pain shows in the details.
Curious how this actually handles sync when two sessions are mid-execution at the same time — is it message queue based or just polling? Also wondering if it stays entirely local or if any coordination data ever leaves my machine.
love how lightweight this stays, no bloated dashboard or extra ceremony, just the agents talking to each other cleanly. the fact that it came out of real workflow pain shows in the details.