How to make one Agent distribute tasks to another? (NOT sub-agents)
I currently have 2 Agents in OpenClaw (example: Bob + Dred) with separate workspaces and different models.
What I want:
- One agent (Bob) acting as an orchestrator / project manager
- Bob should delegate tasks to another agent (Dred)
- Dred executes the task
- Bob receives the result and integrates it
However, in the UI they appear only as independent Agents, not sub-agents.
So I'm trying to understand the correct OpenClaw architecture for this.
---
1. Is the intended way to do this using session_spawn / sub-agent tool, or is there a recommended way for agents to collaborate while remaining separate UI agents?
2. If
- what is the minimal setup?
- allowlist config?
- example workflow?
3. Where should the delegated agent be defined?
- AGENTS.md?
- openclaw.json?
- workspace configuration?
4. Can this be done while keeping:
- separate memory per agent
- different Ollama models/ports
- clean logs showing delegation → execution → return?
---
- Linux
- Local Ollama multi-server setup
- CPU → 11434 (PM / routing agent)
- GPU0 → 11435
- GPU1 → 11436
- Each agent has its own workspace and identity files
---
Goal:
Team-style workflow → PM Agent distributes tasks to specialized Agents, without merging them into a single sub-agent hierarchy.
Any recommended pattern or example setup would be appreciated
What I want:
- One agent (Bob) acting as an orchestrator / project manager
- Bob should delegate tasks to another agent (Dred)
- Dred executes the task
- Bob receives the result and integrates it
However, in the UI they appear only as independent Agents, not sub-agents.
So I'm trying to understand the correct OpenClaw architecture for this.
---
Questions
1. Is the intended way to do this using session_spawn / sub-agent tool, or is there a recommended way for agents to collaborate while remaining separate UI agents?
2. If
session_spawn- what is the minimal setup?
- allowlist config?
- example workflow?
3. Where should the delegated agent be defined?
- AGENTS.md?
- openclaw.json?
- workspace configuration?
4. Can this be done while keeping:
- separate memory per agent
- different Ollama models/ports
- clean logs showing delegation → execution → return?
---
Environment
- Linux
- Local Ollama multi-server setup
- CPU → 11434 (PM / routing agent)
- GPU0 → 11435
- GPU1 → 11436
- Each agent has its own workspace and identity files
---
Goal:
Team-style workflow → PM Agent distributes tasks to specialized Agents, without merging them into a single sub-agent hierarchy.
Any recommended pattern or example setup would be appreciated
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
134,723Members
Resources
Similar Threads
Was this page helpful?
