Launching today
Swarm Studio
Rehearse your launch on 150 AI agents before you ship
2 followers
Rehearse your launch on 150 AI agents before you ship
2 followers
Drop in a launch plan, price change or policy draft and ask one question. Claude maps who is involved, casts up to 150 agents and lets them post and argue on a mock microblog and forum. You get a report with charts, the loudest threads and an analyst write-up, then you can interview any agent. Free and open source. It runs as a private artifact on your own Claude plan, with no API key and no server.
Free
Launch Team
Atlassian LoomTalk it through in Loom. Ship with AI
Promoted
Hey Product Hunt 👋
Before a launch, I used to guess how people would react. Now I rehearse it first.
Swarm Studio is a free, open-source tool I built on Claude. You drop in a plan, ask one question and press start. Here's what happens next:
Claude maps every person and company in your doc
It casts up to 150 agents, from the key players to the general public
They post, quote and argue on a mock microblog and forum, round by round
Every agent is polled at the end, and you get a report with charts, the loudest threads and an analyst write-up
Then you can interview any agent and ask why they pushed back
My first real run: 150 agents, 30 rounds, 821 posts and comments. It surfaced the one objection my plan never answered.
Setup is one paste into Claude Code:
Claude publishes a private copy to your own Claude account. No API key, no server, nothing to host. After that it's a link you open whenever you need it. A 150-agent run takes about 25 minutes on your Claude plan.
Who it's for: founders testing a price change, marketers testing a launch message, agencies showing a client how their audience may react, and product teams checking if anyone cares before they build.
One honest caveat: it's one plausible run, not a forecast. Treat it as a rehearsal and run it more than once.
It's MIT licensed, and it was inspired by MiroFish and rebuilt from scratch to run on Claude.
What would you rehearse first? I'd love your feedback on the report: what would make it more useful?