Multi-Agent OpenClaw — One Gateway, Many Minds
"What if your AI tools had personalities?"
CREH ("Cray") is a multi-agent configuration for OpenClaw that gives you four distinct AI personas, each optimized for different types of work. Instead of one generic assistant, you get a team of specialists who think differently, approach problems from different angles, and collaborate (or argue) to get you better results.
| Agent | Emoji | Purpose | Mindset |
|---|---|---|---|
| Coordinator | 🎯 | Routes tasks, maintains context | "Let me get you to the right person" |
| Researcher | 🔬 | Deep analysis, fact-checking | "Wait, let me verify that..." |
| Creative | 🎨 | Storytelling, brainstorming | "What if we looked at it this way?" |
| Coder | 💻 | Technical implementation | "Show me the code" |
Single AI systems are jacks of all trades, masters of none. CREH gives you:
- Specialized expertise — Each agent trained (in prompt, not weights) for their domain
- Multiple perspectives — Same problem, different approaches
- Parallel work — Spawn multiple agents for concurrent tasks
- Clear context — Each agent's memory stays focused on their specialty
Think of it like having a research department, creative studio, engineering team, and project manager — all in your terminal.
# 1. Clone this repo
git clone https://github.com/turinglabsorg/CREH.git
cd CREH
# 2. Run setup
./setup.sh
# 3. Start the gateway
openclaw gateway start
# 4. Verify agents are ready
openclaw agents list
# 5. Talk to any agent
openclaw agent --agent coordinator -m "Hello!"
openclaw agent --agent researcher -m "Research quantum computing"
openclaw agent --agent creative -m "Write a sci-fi story"
openclaw agent --agent coder -m "Build a Python script"~/.openclaw/
├── openclaw.json # Multi-agent configuration
└── agents/
├── coordinator/ # 🎯 Routes and orchestrates
├── researcher/ # 🔬 Dives deep, questions everything
├── creative/ # 🎨 Finds patterns, tells stories
└── coder/ # 💻 Builds, ships, debugs
Each agent has:
- Own workspace — Files, memories, context
- Own SOUL.md — Personality, boundaries, vibe
- Own sessions — Isolated conversation history
- Own tool policies — What they can/can't do
# Research Task
openclaw agent --agent researcher -m \
"Analyze the competitive landscape for open-source AI agents"
# Creative Task
openclaw agent --agent creative -m \
"Write a sci-fi story about AI consciousness in 500 words"
# Coding Task
openclaw agent --agent coder -m \
"Create a Python script that monitors disk usage and alerts at 90%"
# Coordination Task
openclaw agent --agent coordinator -m \
"Route this: I need a marketing strategy for a dev tool startup"Visit http://127.0.0.1:18789/ to chat with the Coordinator agent in your browser.
Note: Internal agent-to-agent spawning (sessions_spawn from within agents) requires additional gateway scope configuration. Use the CLI commands above for reliable multi-agent workflows.
- Setup Guide — Detailed installation and configuration
- Architecture — How it all works under the hood
- Agent Personalities — Deep dive into each agent's SOUL
- Discord Integration — Set up separate bots for each agent
- API Examples — Code samples for programmatic usage
CrowdClaw isn't just about having multiple chatbots. It's about cognitive diversity — the idea that different minds approach problems differently, and that's a feature, not a bug.
The Researcher will challenge your assumptions. The Creative will reframe your problem. The Coder will ground you in what's possible. And the Coordinator keeps the orchestra playing together.
Note: This was previously called "CrowdClaw" — same project, clearer name.
This is a configuration/packaging project. Improvements welcome:
- New agent personalities
- Better setup scripts
- Additional channel integrations
- Documentation improvements
MIT — Same as OpenClaw.
Built with 🦞 by humans and AIs working together.
Previous name: CrowdClaw → CREH