With codex reducing the usage they give at their 200 usd plan, I had to figure out how to make the agentic coding more efficient. One of the old methods I always knew of but for a while did not have to use was spawning a stupider agent in another terminal and letting Astra plan and Luna execute. this was cheaper, and faster. Issue was that I had to be "meat proxy" - relay messages between the guys at all times.
I am lazy, and i dont like doing stuff. I am more of a manager, visionary, you know? Whatever, I made this repo that lets all agents to chat among each other. Similar to how grokbots agents (idk if you tried it, its cool) chat among each other - but this is model agnostic. Connect your agents to an MCP and whether it is claude or gpt or deepseek - they all can cooridnate in a similar format.
We use inbox and bg polling every n minutes. its fully offline (local). It is still work in progress, and we are working on adding multi-machine controls (online) as well as groupchats with dynamic hierarchies - isntead of X is leader, Y is worker, we can have them move around freely.
nurgaliev•46m ago
I am lazy, and i dont like doing stuff. I am more of a manager, visionary, you know? Whatever, I made this repo that lets all agents to chat among each other. Similar to how grokbots agents (idk if you tried it, its cool) chat among each other - but this is model agnostic. Connect your agents to an MCP and whether it is claude or gpt or deepseek - they all can cooridnate in a similar format.
We use inbox and bg polling every n minutes. its fully offline (local). It is still work in progress, and we are working on adding multi-machine controls (online) as well as groupchats with dynamic hierarchies - isntead of X is leader, Y is worker, we can have them move around freely.
Thank you