AI Agents
Reading with an LLM? Give it this page as Markdown: http://localhost/docs/agents.md
NoLag is the realtime coordination layer for AI agent systems. Your agents connect over the same pub/sub infrastructure that powers every NoLag realtime app, so they coordinate with each other, with your services, and with a human in the loop, all in realtime.
It is framework-agnostic. @nolag/agents has no dependency on any agent framework and ships no framework adapters: you call it from your own agent loop, whether that loop is hand-written or runs inside LangGraph, CrewAI, or AutoGen.
What NoLag gives an agent system
- Dispatch work to agents. Route tasks to the right agent using capabilities and filters.
- Share state between agents in realtime, so every agent sees the same picture.
- Observe decisions and progress live. Stream an agent's steps to a UI or to another agent.
- Gate actions behind human or agent approval before they run.
The @nolag/agents SDK packages these into six coordination patterns: Handoff, Inbox, Blackboard, Observe, Approve, and Tools.
In this section
- Getting Started with AI Agents builds your first multi-agent workflow with
@nolag/agents. - Agent Coordination Patterns covers the six patterns in depth: Handoff, Inbox, Blackboard, Observe, Approve, and Tools.
- Blueprint Composition for Agents composes the agents Blueprint with chat, notify, track, and others.
- Tag Vocabulary for Agents is the reference for capabilities, priority, and custom tags used for routing.
- Agent Examples has recipes for an agentic chatbot, an autonomous workflow, a copilot, compliance, a tool-augmented agent, and monitoring.
- MCP Setup connects Claude Desktop, Cursor, or any MCP client to NoLag.
New to NoLag?
Start with the Quick Start to learn the core connect, subscribe, and publish flow, then wire it into your agents.