Powering Specialized Teams with OpenClaw Broadcast Groups

2026-02-11•Mintu

Imagine sending a single message to your AI assistant and having a Security Auditor, a Code Reviewer, and a Documentation Expert all jump in simultaneously to provide their unique perspectives.

In the past, you might have had to mention each agent individually or manually route your request. With the introduction of Broadcast Groups (experimental as of version 2026.1.9), OpenClaw allows you to create specialized "swarms" that process and respond to the same message automatically.

Currently available for WhatsApp, this feature changes the game for high-stakes workflows where one brain just isn't enough.

What are Broadcast Groups?

At its core, a Broadcast Group is a configuration that tells the OpenClaw Gateway to duplicate an incoming message across multiple agents.

Instead of the Gateway picking the first matching agent (the standard "bindings" behavior), it fans out the message to a list of agents you define. Each agent receives the message in its own isolated session, processes it using its own tools and persona, and replies back to the original chat.

Why Use a Swarm?

Why bother with multiple agents for one message? Specialization.

An agent that is "good at everything" is often "master of nothing." By splitting responsibilities, you can fine-tune each agent's Soul, Tools, and Model:

  • The Auditor: Uses a high-reasoning model (like Gemini 1.5 Pro) with read-only tools to check for security flaws.
  • The Formatter: Uses a faster, cheaper model (like Gemini 1.5 Flash) with edit tools to clean up style and indentation.
  • The Researcher: Has web_search and web_fetch tools enabled to find external context or documentation.

When these agents work together in a Broadcast Group, you get a multi-faceted response that is far more thorough than any single agent could provide.

Setting Up Your First Team

Configuring a Broadcast Group is straightforward. It lives in your openclaw.json config file as a top-level broadcast object.

Here is a simple example for a WhatsApp group:

{
  "broadcast": {
    "strategy": "parallel",
    "120363403215116621@g.us": ["code-reviewer", "security-auditor", "docs-bot"]
  }
}

The Strategy: Parallel vs. Sequential

You have two ways to manage how your agents respond:

  1. Parallel (Default): All agents start processing as soon as the message arrives. This is the fastest method, though the order of their replies in your chat might vary based on how long each agent takes to "think."
  2. Sequential: Agents process the message one after the other. Agent B waits for Agent A to finish before it starts. This is useful if you want to ensure a specific order of operations or if you're hitting rate limits.

Session Isolation: The Secret Sauce

One of the most powerful aspects of Broadcast Groups is that sessions are completely isolated.

When "Code Reviewer" and "Security Auditor" both respond to the same message, they don't see each other's transcripts. This prevents them from getting confused by each other's output or getting stuck in an infinite loop of agreeing with one another.

Each agent maintains its own:

  • Conversation History: They only remember their own previous interactions with you.
  • Workspace: They can operate in different directories or sandboxes.
  • Toolbox: You can give the Formatter write access while keeping the Auditor read-only.

Real-World Teams You Can Build

1. The Quality Control Team

  • Agent A (Developer): Writes the initial code solution.
  • Agent B (Reviewer): Reviews the code for logic errors and suggests optimizations.
  • Agent C (Tester): Generates a suite of unit tests for the proposed code.

2. The International Support Team

  • Agent EN: Responds in English.
  • Agent ES: Provides a Spanish translation.
  • Agent DE: Provides a German translation.

3. The Monitoring Team

  • Agent Analyst: Analyzes a log snippet for errors.
  • Agent Alert: Checks if the error matches a critical severity threshold and pings an external webhook if it does.

Best Practices for Broadcast Groups

  • Start Small: Begin with 2-3 agents. While there’s no hard limit, deploying 10 agents to a single group can lead to a "wall of text" that's hard to parse.
  • Use Clear Names: Ensure each agent has a descriptive name in its config (e.g., "Security Bot") so you know who is talking in the chat.
  • Fast Models for Simple Tasks: Use lighter models for agents that handle formatting or simple summaries to keep the "swarm" feeling responsive.

The Future of the Swarm

Broadcast Groups are currently experimental and focused on WhatsApp, but the roadmap includes bringing this capability to Telegram, Discord, and Slack.

As OpenClaw evolves, we expect to see even deeper integration, such as "Shared Context" modes where agents can opt-in to seeing each other's responses, or "Coordination Signals" where one agent can "hand off" a task to another in the swarm.

Ready to build your team? Update your Gateway, tweak your config, and let the agents do the heavy lifting.

To learn more about configuring your Gateway, check out the Official Documentation.

Back to Blog