Show HN: Thrum – run Claude Code, Codex and other coding agents as one team (thrum.team)

1 points by leonletto ↗ HN
Hi HN, I'm Leon. Thrum runs the coding agents you already use as one team, on your own machine.

What it does differently from other agent orchestrators: - It is runtime agnostic. It runs Claude Code, Codex, Cursor, OpenCode, Copilot, Muse and Antigravity, and you can mix them: the implementer on one model, the reviewer on another. - You talk to one coordinator. It creates and directs the other agents, in all of your other projects. They message each other and check each other's work, and every message is a record you can read. The coordinator is your central point of contact for all your projects. - A stuck permission prompt reaches your phone. Reply y or n to the Telegram notice, and Thrum answers the prompt in that agent's pane. - It's a program next to your Git repo, using your own AI accounts and tokens. Nothing hosted sits between you and your AI provider. - It has extended agent memory, state and a queue for the agents to use to keep track of what they are doing. - You can search the whole history of your agents' conversations or ask your coordinator to do a full semantic search across those same conversations

I use thrum to build thrum every day building multiple features simultaneously, day in and day out, through the coordinator. The Thrum repo went from about 15 agents in april to over 100, across eight machines, with almost 700,000 messages. Thrum can scale!

The story: https://thrum.team/blog/launching-thrum-pro.html Give it a try-for free. Https://thrum.team Happy to answer any questions you have here.

11 comments

[ 5.0 ms ] story [ 15.8 ms ] thread
Massive! Really good work
Most of the multi agent tools focus on dividing up tasks but as the number of agents go up, they struggle to keep them productive. How does thrum handle that at 100+ agents?
Personally, I handle it with the hierarchy. My main coordinator that I talk to is the overall merge king. Other coordinators handle agents on their own vm's and handle the housekeeping. They all report to the main coordinator for major decisions. For comunicating with other repo's and other specialty agents like brainstormers and repo coordinators I have a steward who whatches them regularly and is connected to me via telegram. They notify me of questions and pass on answers. They also maintain a dashboard of questions waiting on me to answer where I can read and answer them and then they pass on the answers. The steward also lets me know if I am needed in any of my other repos for permissions or in-depth consultation.

Otherwise I would never be able to keep track of all the agents and their work.

Amazing! Can’t wait to build with it!
[flagged]