How to Build Multi-Agent Systems in TypeScript

A practical multi-agent layout looks like this:

User
 ↓
Coordinator Agent
 ├── Research Agent
 ├── Support Agent
 ├── Billing Agent
 └── Operations Agent

With HazelJS Agent OS you get:

1. Identity — Agent DNA + in-process registry
2. Communication — @Delegate, supervisor routing, optional A2A
3. Execution — shared AgentRuntime and durable run stores
4. Authorization — Gatekeeper on every tool, including cross-agent paths
5. Observability — timelines via hazel logs / inspector

Start simple

Register agents on one runtime, then introduce a coordinator that delegates. Add Gatekeeper before money-moving tools. Promote packaging to DNA when you need versioned apply.

Read next

• TypeScript multi-agent framework
• Agent orchestration
• A2A guide