Multi-Agent Swarm
A planner decomposes your goal into independent subtasks; worker agents run them in parallel across different models; a judge synthesizes and critiques the result. Watch the orchestration happen live.
How it works
- Planner (gpt-5.4) splits the goal into 2–3 parallelizable subtasks.
- Workers run concurrently — each subtask assigned to a different model (gpt-5.4 / grok-4.3).
- Judge merges the workers' outputs into one answer and critiques the collaboration.
- Every state transition streams over SSE; nodes light up as agents start and finish.
Plannerdecompose
→
Workers ×Nparallel · multi-model
→
Judgesynthesize + critique