praxis-swarm

First Attempt at an Agent Swarm Was interested in saving tokens using Codex and wanted a way to plan with a high model, like Astra, and delegate the work to a lower model, like Luna, for token management. I also wanted to have the work done quickly, so I wanted to have multiple agents work on a task in parallel. Running parallel agents can be a pain though. A lot of context can get dropped, the agents can overwrite each other, etc. I needed a good way to hand off tasks to the sub agents, have the sub agents work in their own work tree, and successfully concatenate the work together. Praxis Swarm is my attempt at that. I built off Agent Orchestrator (Apache 2.0), and added more structure for agent teams (ex: what agent works as a lead vs workers), task scope planning, job assignment packets to sub agents, as well as job freezes and reviews from the lead agent. Definitely not at Kimi Agent Swarm yet! But I’ve tested it with two agents running in parallel with good results! Would love any feedback!