OpenClaw Multi-Agent Workflow: Run Multiple AI Agents on One Instance
You want a Slack bot for customer support and a Telegram assistant for personal tasks. Two agents, two personas, same server. Setting up multi-agent on bare metal means wrangling agentDir configs and binding rules by hand.
This guide covers three patterns: bindings-based routing (one Gateway, multiple agents), sub-agents (spawning workers mid-conversation), and separate instances (full isolation). Plus the Lobster workflow engine for deterministic YAML pipelines where each step pipes JSON to the next.
Biggest gotcha: never reuse an agentDir between agents. Session collisions where Agent A answers with Agent B's personality. Trust me on this one.
https://clawhosters.com/blog/posts/openclaw-multi-agent-workflow
Top comments (0)