SW
OpenCode Direct API Multi-agent swarm experiment
sandbox

Multi-agent swarm

Several agents work simultaneously on one prompt. Every agent streams its reasoning and its message live, and each agent reads the shared team log (what everyone else already did and thought) before contributing — so they build on each other instead of duplicating each other. A final synthesizer merges everything.

Set up the swarm

Team board — live reasoning and messages, all at once

Configure the swarm above and press Launch.

Shared team log — what every agent reads before speaking

Nothing yet. The team log grows as the swarm works.

This is a sandboxed client-side experiment. The API key is only sent to this page's same-origin proxy and straight to opencode, never stored server-side. Agents read each other through a shared team log that is appended after every round; within a round all agents stream in parallel.