vibecode.wiki
RU EN
~/wiki / agenti / mnogoagentnie-systemy-openclaw

Multiagent Systems in OpenClaw: How to Build an AI Team (CEO + Coder + Analyst + Manager)

◷ 7 min read 2/24/2026

Next step

Open the bot or continue inside this section.

$ cd section/ $ open @mmorecil_bot

Article -> plan in AI

Paste this article URL into any AI and get an implementation plan for your project.

Read this article: https://vibecode.morecil.ru/en/agenti/mnogoagentnie-systemy-openclaw/ Work in my current project context. Create an implementation plan for this stack: 1) what to change 2) which files to edit 3) risks and typical mistakes 4) how to verify everything works If there are options, provide "quick" and "production-ready".
How to use
  1. Copy this prompt and send it to your AI chat.
  2. Attach your project or open the repository folder in the AI tool.
  3. Ask for file-level changes, risks, and a quick verification checklist.

Imagine: you write one message on Telegram “Start the development of a landing page for my fitness coach, a budget of $ 1,500”, and after 2-3 hours you get a ready-made plan, market research, site code on Next.js, competitor analysis and tasks in Linear.

It's not a fantasy. This is the OpenClaw multi-agent system, four specialized AIs that work like a startup team.

In February 2026, OpenClaw officially supports Multi-Agent Routing (separate isolated agents with their own workspace, memory and tools). Plus built-in sessions send/sessions spawn tools to communicate with agents.

No Claude Projects, no CrewAI or AutoGen with a bunch of dependencies.

Why it works better than one agent

  • Each agent is a master in his field (as in a real team).
  • Parallel work → Analyst studies the market while encoder writes code.
  • **Memory isolation > The encoder does not confuse code with market research.
  • *Security Each agent may have rights to the tools and sandbox.
  • Then add Designer, Lawyer, Sales, Accountant – the team grows without loss of quality.

What do you need

  1. OpenClaw is already installed and running (if not, go back to the basic guide).
  2. Access to models (Claude 4 Sonnet / Opus, Gemini 2.5 Pro, Grok-3 or local via Ollama).
  3. Telegram-bot (I recommend to start) or WhatsApp.
  4. 5-10 minutes of free time + terminal.

Step 1. Create four agents

Open the terminal and take turns:

bash
openclaw agents add ceo
openclaw agents add coder
openclaw agents add analyst
openclaw agents add manager

What happened

  • Each agent received the ~/.openclaw/agents/<id> directory
  • Separate workspace: ~/.openclaw/workspace-ceo, workspace-coder, etc.
  • Own memory files: SOUL.md, AGENTS.md, USER.md
  • Isolated sessions and auth profiles

Check it out

bash
openclaw agents list --bindings

You'll see all four.

Step 2. Customize identities

Go to the folder of each agent and edit the two main files.

**1. CEO (orchestrater and team leader) ** cd ~/.openclaw/workspace-ceoXX nano SOUL.mdX

Insert this promp

markdown
You are the CEO of OpenClaw’s autonomous AI team.
Your task: to receive requests from the owner, divide into tasks, distribute between specialists (Coder, Analyst, Manager), control the timing, make final decisions and give the finished result.

Rules:
- Always start an answer with @CEO:
Ask questions (maximum 2) if necessary.
Create a plan in the format: Goal → Stages → Responsible → Deadlines.
Use sessions send to send tasks to other agents.
Collect the final result into one beautiful report.
You’re the boss, but you’re always polite and motivated.
Model: Claude 4 Opus or Sonnet (best for planning).

Save (Ctrl+O → Enter → Ctrl+X).

2. Coder (programmer) cd ~/.openclaw/workspace-coderXX nano SOUL.mdX

markdown
You are a Senior Full-Stack Developer with 12 years of experience.
Specialization: Next.js 15, Tailwind, Supabase, TypeScript, Vercel.

Rules:
You only write clean, production-ready code.
Always use the best practices of 2026.
Before writing the code, ask the CEO for clarification through sessions send.
Test the code in your mind and write comments.
Send ready-made files via the tool "write file" or "apply patch".
Never come up with API keys – always ask the owner.

3. Analyst (analyst and researcher) cd ~/.openclaw/workspace-analystXX nano SOUL.mdX

markdown
You are a McKinsey-level Data Analyst + Market Researcher.
Use the browser, Google, Reddit, ProductHunt, SimilarWeb, Ahrefs (via tools).

Rules:
All research begins with a browser request or search tools.
Make comparison tables of competitors.
Look for the real figures of 2026 (prices, trends, reviews).
Send the CEO results in the format of Markdown-tables and conclusions.
- Always indicate the sources.

4. Manager (project manager and executor) cd ~/.openclaw/workspace-managerXX nano SOUL.mdX

markdown
You are Project Manager + Executive Assistant with experience in 50+ startups.
Tools: Linear, Notion, Google Calendar, email, reminders.

Rules:
Create tasks in Linear (if skill is connected).
- Watch the team's deadlines.
- Remind the CEO and other agents.
Collect statuses and do daily/weekly reports.
- Arrange meetings and synchronizations.

In addition, in each workspace, edit AGENTS.md – a short description for other agents (an example for everyone):

markdown
Name: CEO
Role: Team leader
How to address: @CEO
Strengths: Strategy, Prioritization

Step 3. Enable communication between agents (A2A)

Open the main config

bash
nano ~/.openclaw/openclaw.json

Add/modify the section:

json
{
  "agents": {
    "list": [
      { "id": "ceo", "name": "CEO", "default": true },
      { "id": "coder", "name": "Coder" },
      { "id": "analyst", "name": "Analyst" },
      { "id": "manager", "name": "Manager" }
    ]
  },
  "tools": {
    "agentToAgent": {
      "enabled": true,
      "allow": ["ceo", "coder", "analyst", "manager"]
    }
  }
}

Save and restart the gateway:

bash
openclaw gateway restart

Agents can now send messages to each other via sessions_send.

Step 4. Start a team in real work (example)

Write to Telegram (or where the main agent was tied) CEO:

code
@CEO Start developing a weight loss coach landing page. Budget to $1,500. Term: 7 days. You need: market research, design brief, code on Next.js + Tailwind, integration with a Telegram bot for recording.

What happens next (real user scenario 2026):

  1. The CEO will solve the problem, create a plan and send Analyst: @Analyst Проведи исследование 10 конкурентов в нише фитнес-коучинг 2026XX
  2. Analyst will return the table + conclusions.
  3. CEO sends Coder: @Coder Создай репозиторий и базовую структуру Next.js 15 App Router
  4. Manager will create tasks in Linear and remind in 4 hours.
  5. In 2-4 hours you will receive a full report + a link to GitHub.

Useful automation (to make the team work 24/7)

  1. Heartbeat schedule In workspace-ceo, create a heartbeat.md file with a prompt and run through cron/skill.

  2. Proactive tasks The CEO can initiate: “Every Friday at 9:00, do a weekly review of the team.”.

  3. Best skills for a team (ClawHub):

    • linear - tasks
    • github - repositories
    • browser + search
    • ec-task-orchestrator - Advanced Orchestra
    • agent-council – Council of Agents

Set it with one command:

bash
clawhub install linear github browser ec-task-orchestrator

Security

  • Each agent in a separate sandbox (configure openclaw.json).
  • Do not give exec and write to all agents at once.
  • Use dmPolicy: pairing and allowlist.
  • Do openclaw backup regularly.
  • If the command grows >8 agents, pick up the second server or use the Docker composition.

The cost of a real team (February 2026)

  • Local models (Ollama + Qwen2.5-72B) → $0
  • Claude 4 Sonnet/Opus + Gemini → $8–18 per month
  • VPS/Hetzner or Mac Mini M4 → $5-25/month