Agent Orchestrator is a framework for building multi-agent coding systems.
It handles the complexity of coordinating multiple AI agents—planning,
execution, review—so you can focus on the actual problem.
The key insight: most coding tasks benefit from specialization. A planning
agent thinks differently than an execution agent. By separating concerns,
we get better results than a single monolithic agent.
import { Orchestrator, PlanningAgent, ExecutionAgent } from 'agent-orchestrator';
const orchestrator = new Orchestrator({
agents: [
new PlanningAgent({ model: 'gpt-4' }),
new ExecutionAgent({ model: 'gemini-pro' }),
],
routing: 'auto',
});
const result = await orchestrator.run({
task: 'Refactor the authentication module to use JWT',
context: { codebase: './src' },
});
Agent coordination is hard. The biggest challenge wasn't the AI—it was
managing state and handling failures gracefully. Redis turned out to be
essential for distributed agent state.