Upvote this launch
Product media preview
5 launch images
About this launch
Work goes in from Slack, GitHub or the API, and a reviewed pull request comes back to your repo. A lead agent splits the goal into tasks and hands them to coding agents such as Claude Code or Codex. Every agent runs in its own Docker container. The team keeps what it learns: shared memory, reusable skills, schedules, and a review before anything merges. Work survives a restart. It is MIT and self-hosted, and runs on your own model keys. 940+ stars on GitHub.
Hi, I'm Taras, co-founder and CTO of Desplega Labs. We built Agent Swarm because one agent stopped being enough for our own work: research, plans, code, reviews and the follow-ups each needed different tools and a different check. So we made it a factory you run yourself. Work comes in from Slack, Linear, email or GitHub, and a lead agent turns it into tasks for coding agents like Claude Code or Codex. Your team adds the steps it cares about, like tests, a review or a risk check. Low-risk changes can then merge on their own, and risky ones wait for a person. It's MIT and runs on your own infrastructure and model keys. We run it on our own repo every day, and Capchase runs it with +60 active users. Ask me anything about the review gates, what it costs to run, or what breaks when you give agents real work.