Can OpenClaw run multiple agents simultaneously?

Yes — agents run in parallel via the same gateway instance, routed by trigger phrase or skill context. Each agent has its own system prompt and skill set; they share the same gateway and API keys without interfering with each other.

How many agents can I run on GetClaw?

Starter supports 1–2 active agents, ideal for solo founders. Team supports up to 5 agents with shared gateway and multi-user approval flows. Managed Plus is unlimited — no cap on concurrent agents or team members.

Do I need separate API keys per agent?

No — BYOK means one API key set, all agents use it with no GetClaw markup. You connect your Anthropic, OpenAI, or other provider keys once; every agent in your team routes through the same credentials at cost.

Can different team members trigger different agents?

Yes — GetClaw's permission system lets you assign agents to team members. A writer on Slack can trigger @write and @review but not @outreach. A team lead can trigger all agents. Permissions are set per user from the GetClaw dashboard.

How do I prevent agents from sending emails without approval?

OpenClaw's approval controls flag sensitive skills — outreach actions require explicit approval before executing. You define which skills need approval in your openclaw.yml routing config; the agent queues the action and notifies the designated approver instead of firing immediately.

Team Setup Guide · March 2026 · 10 min read

OpenClaw Multi-Agent Setup:
Run Like a Team of 10

One gateway. Multiple specialised agents. Replace your research, writing, review, and outreach stack — from $29/mo.

What multi-agent OpenClaw actually looks like

Most people run one OpenClaw agent — a general assistant they ping when they need something. Teams that get real leverage from OpenClaw run 4–8 specialised agents in parallel via one gateway, routing tasks based on context. Instead of one agent that does everything passably, you have a Research Analyst that goes deep on sources, a Writer that produces clean copy, and a Reviewer that catches errors before they ship — all triggered from the same Slack channel, all on the same gateway, all using your own API keys with zero markup.

The architecture is straightforward: each agent has a dedicated system prompt that sharpens its role, a curated set of skills matched to its job, and a trigger phrase that routes messages from your team. A common team stack covers five roles — Researcher, Writer, Reviewer, Outreach, and Scheduler. Together they handle the full lifecycle of a content or outreach workflow without a single additional SaaS subscription. All through one GetClaw gateway instance.

The "Team of 10" agent stack

Research Analyst

Deep-dives topics on command, summarises findings, fact-checks claims before they reach a writer.

Brave SearchFirecrawlWikipedia

@research [topic]

Content Writer

Writes blog posts, emails, and proposals from research summaries — with consistent voice and structure.

MemoryNotion writeGoogle Docs

@write [brief]

Quality Reviewer

Critiques drafts, checks claims against research, flags inconsistencies and factual errors before publishing.

MemoryFile readBrave Search

@review [document]

Outreach Agent

Drafts personalised emails, LinkedIn messages, and follow-ups — with approval required before any send.

Gmail sendLinkedIn (if permitted)Memory

@outreach [contact]

Project Scheduler

Tracks tasks, sets reminders, and updates project docs so nothing falls through the cracks.

Google CalendarNotionSlack notify

@schedule [task]

Routing configuration

OpenClaw's routing config lives in a single openclaw.yml at the root of your gateway. Each agent gets a system prompt block, a skill list, and a trigger. Sensitive skills — like outreach and scheduler — are set to require explicit approval before firing.

openclaw.yml

# openclaw.yml — multi-agent routing
agents:
  research:
    system_prompt: "You are a research specialist. Deep-dive
      topics, summarise findings clearly, cite every source."
    skills: [brave_search, firecrawl, memory]
    trigger: "@research"

  writer:
    system_prompt: "You are a content writer. Produce clean,
      structured copy from research summaries. Match brand voice."
    skills: [memory, notion, google_docs]
    trigger: "@write"

  reviewer:
    system_prompt: "You are a critical editor. Review drafts
      against source material. Flag errors, gaps, and weak claims."
    skills: [memory, brave_search]
    trigger: "@review"

  outreach:
    system_prompt: "You are an outreach specialist. Draft
      personalised messages. Never send without approval."
    skills: [gmail_send, memory]
    trigger: "@outreach"

  scheduler:
    system_prompt: "You are a project scheduler. Track tasks,
      set reminders, update docs. Confirm before all calendar writes."
    skills: [google_calendar, notion, slack_notify]
    trigger: "@schedule"

routing:
  default_agent: research
  approval_required: [outreach, scheduler]

Real workflow: blog post in 22 minutes

The same five-agent stack handles the full lifecycle — from research to final draft — with two humans in the loop at key approval points.