Live SimulationMulti-Agent·Live Engine
Autonomous Multi-Agent Telemetry & Delegation Simulator
Real-time state machine visualizer demonstrating supervisor-to-worker dispatching, reasoning pipelines, and packet telemetry.
Live Hardware-Accelerated ViewportDrag to rotate · Interact with controls
System Architecture & Engineering Concept
Simulate how modern autonomous agent runtimes (like KendaliAI) delegate work between specialized LLM agents. Features a central Coordinator Agent dispatching structured JSON task packets to Coder, Security Auditor, and Architect worker agents with live state transitions and event streams.
Mathematical Formulation / Heuristic:
Directed Acyclic Graph (DAG) state progression: G = (V, E), where V is the set of agent nodes and E is the message passing latency vector.
Implementation Highlights
- Event-driven reactive state machine in pure React with animated SVG telemetry pipelines.
- Live message ledger recording token throughput, prompt payload size, and subagent roundtrip latency.
- Manual error injection and retry policy simulation demonstrating self-healing agent loops.
Interactive Parameters
Trigger Task
Dispatch a synthetic engineering prompt through the agent coordinator DAG
Agent Speed
Adjust simulated inference latency (100ms - 2000ms per agent step)
Fault Injection
Inject synthetic subagent timeout to test autonomous supervisor recovery
Core Engine LoopTypeScript
async function orchestrate(task) {
const plan = await coordinator.decompose(task);
const results = await Promise.all(
plan.subtasks.map(st => workerPool.dispatch(st.role, st.payload))
);
return coordinator.synthesize(results);
}