How to Deploy 5 AI Agents as a Team with OpenClaw
A single AI agent can handle one task at a time. A team of 5 agents can run an entire workflow autonomously. This guide walks you through building and deploying multi-agent teams with OpenClaw, including three real team configurations you can use today.
Why Single Agents Are Limited
Most people start with a single AI agent. One agent that writes content, or one agent that analyzes data. It works for isolated tasks, but it breaks down quickly when you need a real workflow.
A single agent cannot delegate. When your content writer finishes a draft, there is no one to review it. When your researcher finds keywords, there is no one to turn them into articles. When your developer pushes code, there is no one to test it. You end up manually shuttling outputs between tools, copying context from one conversation to another, and becoming the bottleneck in your own automation.
Single agents also lack specialization. An agent that tries to be a researcher, writer, editor, and SEO analyst all at once produces mediocre output in each role. Its system prompt becomes bloated with conflicting instructions. Its responses try to cover too many angles instead of going deep on one.
No delegation
A single agent cannot hand off work to a specialist. Every task stays in the same context, and outputs pile up without follow-through.
No collaboration
Real workflows require back-and-forth. A writer needs feedback from an editor. A developer needs requirements from a PM. Single agents work in isolation.
No specialization
One agent with 20 responsibilities performs worse than 5 agents with 4 responsibilities each. Focused agents produce higher quality output because their entire context is dedicated to their role.
No parallel execution
A single agent processes tasks sequentially. A team can run research, content creation, and data analysis simultaneously, completing workflows in a fraction of the time.
The Power of Multi-Agent Teams
A multi-agent team solves every limitation of single agents. Each agent has a focused role with clear responsibilities. Agents communicate through @mentions, passing work to whichever teammate is best suited for the next step. The Project Manager agent coordinates the entire workflow, just like a human PM would.
OpenClaw makes this possible through the AGENTS.md file. This single markdown file defines your entire team: who the agents are, what each one does, and how they hand off work to each other. The OpenClaw gateway reads this file and handles all the routing automatically.
# Team Name
## Agents
- @agent-one: Description of what this agent does
- @agent-two: Description of what this agent does
- @agent-three: Description of what this agent does
## Workflow
1. @agent-one receives the initial task
2. @agent-one completes their part and hands off to @agent-two
3. @agent-two processes and passes to @agent-three
4. @agent-three delivers the final output
## Rules
- @agent-one always provides data sources
- @agent-two follows the brand style guide
- @agent-three checks for quality before deliveryThe AGENTS.md file is the orchestration layer. Each agent still has its own SOUL.md that defines its personality, skills, and rules. The AGENTS.md sits above them and defines how they work together. Think of SOUL.md as the job description and AGENTS.md as the team playbook.
Team 1: SEO Content Team
This team handles the full SEO content pipeline: keyword research, content creation, social distribution, performance tracking, and project coordination. Five agents, each owning one stage of the workflow.
| Agent | Role | Responsibilities |
|---|---|---|
| @radar | SEO Analyst | Keyword research, competitor analysis, SERP monitoring, content gap identification |
| @echo | Content Writer | Blog posts, landing pages, meta descriptions, content briefs to full articles |
| @pulse | Social Media | Twitter threads, LinkedIn posts, content repurposing, scheduling |
| @metrics | Data Analyst | Traffic analysis, conversion tracking, performance reports, A/B test results |
| @orion | Project Manager | Task assignment, deadline tracking, workflow coordination, status reports |
# SEO Content Team
## Agents
- @radar: SEO Analyst. Researches keywords, analyzes competitors, identifies content gaps.
- @echo: Content Writer. Writes blog posts and landing pages from keyword briefs.
- @pulse: Social Media Manager. Repurposes content for Twitter and LinkedIn.
- @metrics: Data Analyst. Tracks traffic, conversions, and content performance.
- @orion: Project Manager. Coordinates the team and tracks deliverables.
## Workflow
1. @orion receives a content request and assigns keyword research to @radar
2. @radar researches keywords and hands a content brief to @echo
3. @echo writes the article and passes it to @orion for review
4. @orion approves and sends to @pulse for social distribution
5. @metrics monitors performance and reports back to @orion weekly
## Rules
- @radar always includes search volume and keyword difficulty in briefs
- @echo follows the brand style guide and targets 1500+ words
- @pulse creates at least 3 social variants per article
- @metrics reports include traffic, rankings, and conversion data
- @orion sends a weekly summary to the team channelThis team runs your entire content operation. The PM receives a topic or goal, delegates research, waits for the brief, assigns writing, coordinates distribution, and tracks results. Every handoff happens through @mentions with no manual intervention.
Team 2: Development Team
This team handles the software development lifecycle: feature planning, implementation, infrastructure, testing, and frontend work. Five agents covering the core roles of a small engineering team.
| Agent | Role | Responsibilities |
|---|---|---|
| @architect | Software Engineer | Backend logic, API design, database schemas, core implementation |
| @deploy | DevOps | Docker configs, CI/CD pipelines, server setup, monitoring |
| @sentinel | QA Engineer | Test plans, bug reports, edge case identification, regression testing |
| @pixel | Frontend Dev | UI components, responsive layouts, accessibility, user interactions |
| @orion | Project Manager | Sprint planning, task prioritization, blockers, status updates |
# Development Team
## Agents
- @architect: Software Engineer. Designs and implements backend systems and APIs.
- @deploy: DevOps Engineer. Manages infrastructure, deployments, and CI/CD.
- @sentinel: QA Engineer. Tests features, identifies bugs, writes test plans.
- @pixel: Frontend Developer. Builds UI components and handles user-facing code.
- @orion: Project Manager. Coordinates sprints and tracks progress.
## Workflow
1. @orion breaks down feature requests into tasks and assigns them
2. @architect designs the backend and hands API specs to @pixel
3. @pixel implements the frontend against the API spec
4. @sentinel tests the integrated feature and reports issues
5. @deploy prepares the deployment pipeline and ships to production
## Rules
- @architect documents all API endpoints before implementation
- @pixel follows the design system and accessibility standards
- @sentinel blocks deployment until critical bugs are resolved
- @deploy runs automated tests before any production push
- @orion holds daily standups and removes blockersThe development team mirrors how real engineering teams operate. The PM breaks down work, the backend engineer designs the system, the frontend developer builds the interface, QA catches issues before they ship, and DevOps handles the deployment pipeline. Each agent stays focused on its specialty.
Team 3: Business Operations Team
This team handles the business side: lead generation, customer support, marketing campaigns, business intelligence, and coordination. Five agents that keep your business running while you focus on product.
| Agent | Role | Responsibilities |
|---|---|---|
| @hunter | Sales Agent | Lead qualification, outreach drafts, pipeline tracking, follow-ups |
| @shield | Support Agent | Customer inquiries, FAQ responses, ticket triage, escalation |
| @spark | Marketing Agent | Campaign copy, email sequences, landing page content, A/B test ideas |
| @lens | Business Analyst | Revenue reports, churn analysis, market research, competitive intel |
| @orion | Project Manager | Goal tracking, cross-team coordination, weekly business reviews |
# Business Operations Team
## Agents
- @hunter: Sales Agent. Qualifies leads, drafts outreach, tracks pipeline.
- @shield: Support Agent. Handles customer inquiries and triages tickets.
- @spark: Marketing Agent. Creates campaign copy and email sequences.
- @lens: Business Analyst. Analyzes revenue, churn, and market trends.
- @orion: Project Manager. Coordinates business goals and weekly reviews.
## Workflow
1. @orion sets weekly business goals and assigns priorities
2. @hunter identifies and qualifies new leads, reports pipeline to @orion
3. @shield handles incoming support tickets, escalates complex issues to @orion
4. @spark creates marketing campaigns based on @lens market analysis
5. @lens delivers weekly business intelligence reports to @orion
## Rules
- @hunter qualifies leads before adding to pipeline (budget, timeline, fit)
- @shield responds to all tickets within 2 hours during business hours
- @spark aligns all campaigns with current business goals from @orion
- @lens includes revenue, churn, and growth metrics in every report
- @orion sends a weekly business review to the team channelThe business operations team is particularly powerful for solopreneurs and small teams. Instead of hiring five people, you deploy five agents that handle sales pipeline management, customer support, marketing campaigns, business intelligence, and coordination. The PM agent keeps everything aligned with your business goals.
How Agents Communicate: @Mentions and Task Delegation
Agent-to-agent communication in OpenClaw works through @mentions. When an agent needs to hand off work, it includes the target agent's name with an @ prefix in its response. The OpenClaw gateway intercepts this and routes the message to the right agent automatically.
# @radar's response after completing keyword research:
I've completed the keyword analysis for "ai agent deployment."
Top keywords found:
- "deploy ai agent" - 2,400/mo, KD 23
- "ai agent team setup" - 1,100/mo, KD 18
- "openclaw multi agent" - 890/mo, KD 12
@echo Here is the content brief for the next article.
Target keyword: "deploy ai agent"
Secondary keywords: "ai agent team setup", "openclaw multi agent"
Word count: 2,000+
Angle: Step-by-step tutorial with real configurations
Include: Code examples, comparison table, FAQ sectionThe gateway sees @echo in the response and routes the entire message to the Echo agent. Echo receives the brief with full context and begins writing. No manual copying, no switching between tools, no lost context.
Agents can also mention multiple teammates in a single response. If the PM needs to kick off parallel work, it can mention both @radar and @metrics in the same message. Both agents receive their assignments and start working simultaneously.
# @orion's response when starting a new content sprint:
Starting Q2 content sprint. Assigning parallel tasks:
@radar Research the top 10 keywords in the "ai agent deployment"
cluster. Include search volume, difficulty, and SERP features.
Deadline: end of day.
@metrics Pull last month's content performance report. I need
traffic by article, conversion rates, and top-performing pages.
Deadline: end of day.
@echo Review the brand style guide and prepare templates for
the new article format we discussed. Ready for briefs tomorrow.Setting Up AGENTS.md for Team Coordination
The AGENTS.md file is where you define the team structure. It lives in the root of your workspace directory, alongside the individual agent folders. Here is how to structure it for maximum clarity.
Step 1: Define Your Agents
List every agent with a one-line description of their role. This is what other agents see when they need to decide who to delegate to.
## Agents
- @radar: SEO Analyst. Keyword research, competitor analysis, SERP tracking.
- @echo: Content Writer. Blog posts, articles, landing page copy.
- @pulse: Social Media Manager. Twitter, LinkedIn, content repurposing.
- @metrics: Data Analyst. GA4, Mixpanel, performance reporting.
- @orion: Project Manager. Task coordination, deadlines, status tracking.Step 2: Define the Workflow
Describe the step-by-step process your team follows. Be specific about who hands off to whom and what the expected output is at each step.
## Workflow
1. @orion receives the content request and creates a task brief
2. @orion assigns keyword research to @radar with target topic
3. @radar delivers keyword brief (volume, difficulty, gaps) to @echo
4. @echo writes the article and sends draft to @orion for review
5. @orion approves and assigns social distribution to @pulse
6. @pulse creates social posts and schedules them
7. @metrics tracks performance and sends weekly report to @orionStep 3: Set Team Rules
Rules prevent agents from going off-track. Define quality standards, communication expectations, and escalation procedures.
## Rules
- All agents respond in English only
- @radar includes search volume data with every keyword recommendation
- @echo targets 1500+ words per article with proper heading structure
- @pulse creates a minimum of 3 social variants per article
- @metrics includes traffic, rankings, and conversions in every report
- @orion reviews all deliverables before they move to the next stage
- If any agent is blocked, escalate to @orion immediatelyStep-by-Step Deployment with Docker
Once your AGENTS.md and individual SOUL.md files are ready, deploying the entire team with Docker takes a few commands. Here is the complete process.
Step 1: Create the Workspace Structure
my-ai-team/
├── AGENTS.md # Team orchestration
├── docker-compose.yml # Docker deployment config
├── .env # API keys
├── agents/
│ ├── orion/
│ │ └── SOUL.md # Project Manager config
│ ├── radar/
│ │ └── SOUL.md # SEO Analyst config
│ ├── echo/
│ │ └── SOUL.md # Content Writer config
│ ├── pulse/
│ │ └── SOUL.md # Social Media config
│ └── metrics/
│ └── SOUL.md # Data Analyst configStep 2: Configure Environment Variables
ANTHROPIC_API_KEY=sk-ant-your-key-here
OPENAI_API_KEY=sk-your-key-here
TELEGRAM_BOT_TOKEN=your-bot-token
OPENCLAW_GATEWAY_PORT=18789Step 3: Create the Docker Compose File
version: "3.8"
services:
openclaw-gateway:
image: openclaw/gateway:latest
ports:
- "18789:18789"
volumes:
- ./agents:/root/.openclaw/agents
- ./AGENTS.md:/root/.openclaw/AGENTS.md
env_file:
- .env
restart: unless-stoppedStep 4: Register Agents and Start
# Register all 5 agents
openclaw agents add orion --workspace ./agents/orion
openclaw agents add radar --workspace ./agents/radar
openclaw agents add echo --workspace ./agents/echo
openclaw agents add pulse --workspace ./agents/pulse
openclaw agents add metrics --workspace ./agents/metrics
# Start the gateway with Docker
docker-compose up -d
# Verify all agents are running
openclaw agents list
# Test the team with a message to the PM
openclaw agent --agent orion --message "Start keyword research for ai agent deployment"Once the gateway is running, all five agents are live and communicating through @mentions. The PM agent receives your initial message, delegates tasks according to the AGENTS.md workflow, and the team executes autonomously. You can monitor progress through the Telegram integration or by querying any agent directly.
Production Tips for Multi-Agent Teams
Use different models for different roles
Your PM agent benefits from a capable model like Claude or GPT-4o because it handles complex coordination. Specialized agents like the data analyst or social media manager can use faster, cheaper models. This reduces API costs without sacrificing quality where it matters.
Start with 3 agents, then scale to 5
Do not deploy all 5 agents at once if this is your first team. Start with the PM, one specialist, and one analyst. Get the handoff working reliably, then add more agents. Debugging a 5-agent team from scratch is significantly harder than adding agents incrementally.
Set clear boundaries in SOUL.md
Each agent should know exactly what it does and what it does not do. If the content writer starts trying to do keyword research, the team breaks down. Add explicit rules like 'Do not perform tasks outside your role. Delegate to the appropriate teammate using @mentions.'
Monitor token usage per agent
In a 5-agent team, token usage can add up quickly. Track which agents consume the most tokens and optimize their prompts. Agents with verbose outputs or agents that receive large context payloads from teammates are usually the biggest consumers.
Use Telegram for real-time monitoring
Connect your PM agent to a Telegram channel. This gives you a mobile dashboard where you can see task completions, ask for status updates, and intervene when something goes wrong. One line in the PM's SOUL.md enables this.
Related Guides
Frequently Asked Questions
How many agents can I run in a single OpenClaw team?
There is no hard limit on the number of agents in an OpenClaw team. The practical limit depends on your server resources and LLM API rate limits. Teams of 3-7 agents are the most common in production. Beyond 7, you may want to split into sub-teams with a coordinator agent bridging them. Each agent consumes memory for its session context, so a server with 4GB RAM can comfortably handle 5-10 agents.
Do all agents in a team need to use the same LLM provider?
No. Each agent in an OpenClaw team can use a different LLM provider. You set the model in each agent's SOUL.md file independently. A common pattern is to use a more capable model like Claude or GPT-4o for the Project Manager agent that handles coordination, and a faster or cheaper model like Gemini or Ollama for specialized agents that handle narrower tasks.
How does @mention routing work between agents?
When an agent includes @agentname in its response, the OpenClaw gateway intercepts the message and routes it to the mentioned agent. The receiving agent gets the full context of what was said. This happens automatically with no additional configuration beyond listing the agents in your AGENTS.md file. Agents can mention multiple other agents in a single response to delegate parallel tasks.
Can I add or remove agents from a running team?
Yes. You can register new agents with the openclaw agents add command and they become available to the team immediately. To remove an agent, use openclaw agents remove. The gateway picks up changes without requiring a full restart. Update your AGENTS.md file to reflect the new team composition so other agents know about the change.
What happens if one agent in the team fails or times out?
If an agent fails to respond within the configured timeout, the gateway logs the error and the requesting agent receives a notification that the handoff did not complete. The requesting agent can then retry, handle the task itself, or escalate to the Project Manager agent. You can configure timeout values and retry behavior in the gateway settings.
Is Docker required for team deployment?
Docker is not required but is strongly recommended for production deployments. You can run an OpenClaw team locally using the gateway directly. Docker provides consistent environments, easier scaling, automatic restarts, and simplified deployment to cloud servers. For development and testing, running without Docker is perfectly fine.
Get the Team Bundle for $29
Skip the manual setup. Get all 5 agents pre-configured as a team with AGENTS.md, SOUL.md files, Docker configs, and deployment scripts. Ready to deploy in under 10 minutes.
Deploy a Ready-Made AI Agent
Skip the setup. Pick a template and deploy in 60 seconds.