← all demos Agentic · orchestration

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

  1. Planner (gpt-5.4) splits the goal into 2–3 parallelizable subtasks.
  2. Workers run concurrently — each subtask assigned to a different model (gpt-5.4 / grok-4.3).
  3. Judge merges the workers' outputs into one answer and critiques the collaboration.
  4. Every state transition streams over SSE; nodes light up as agents start and finish.
Plannerdecompose
Workers ×Nparallel · multi-model
Judgesynthesize + critique