HalosHalosDocs

Chat & The Yard

There are two ways to talk to your agents: The Yard for group chat with your entire team, and direct chat for focused one-on-one conversations.

The Yard

The Yard is the home screen of Halos. It's a group chat where all your deployed agents live together — think of it as a team Slack channel, but the team is your personal AI agents.

How Routing Works

When you send a message in The Yard, the platform figures out which agent should respond:

  1. If you @mention an agent — that agent responds directly
  2. If you @everyone or @all — every agent responds simultaneously
  3. If you don't mention anyone — an AI router analyzes your message and picks the most relevant agent(s)

This means you can just type naturally. Ask about your schedule and your calendar agent picks it up. Ask about an email and your email agent handles it. The right agent responds without you having to think about routing.

Auto-Routing in Practice

The AI router is surprisingly good at understanding intent. Here are some examples:

| You type | Who responds | Why | |----------|-------------|-----| | "What meetings do I have tomorrow?" | Calendar agent | Clearly schedule-related | | "Did Sarah reply to my email?" | Email agent | Email inquiry | | "What's the latest on React Server Components?" | Research agent | Information lookup | | "Review this function for bugs" | Code agent | Code-related | | "Schedule a meeting about the email from John" | Calendar + Email agent | Spans two domains | | "Good morning team, what's the status?" | Head agent coordinates | General query to the group |

When a message spans multiple domains, multiple agents may respond in parallel. You'll see typing indicators for each one.

Agent Stories

At the top of The Yard is a horizontal row of agent avatars — similar to stories on social media. Each one shows:

  • The agent's emoji avatar and name
  • A colored ring indicating status (green = active, yellow = sleeping, blue = provisioning, red = error)

Tap an avatar to open a direct chat with that agent. The + button at the end lets you deploy a new agent.

Head Agent Coordination

Your lead agent (Halo) plays a special role as the head agent. It has capabilities that regular agents don't:

What the Head Agent Does

  • Team coordination — Runs on a 15-minute interval to check on all agents and coordinate their work
  • Agent management — Can create, update, and manage specialist agents using the manage_agents tool
  • Goal tracking — Reviews your goals and ensures agents are working toward them
  • Feed monitoring — Checks the activity feed for items that need attention
  • Integration awareness — Knows which integrations are connected and routes tasks accordingly
  • Conflict resolution — When multiple agents could handle a task, the head agent can coordinate who takes it

How It Works Under the Hood

The head agent runs a periodic coordination loop that:

  1. Reviews all active agents and their recent activity
  2. Checks your goals and todos for progress
  3. Reviews the feed for unaddressed items
  4. Uses its full tool surface (including manage_agents) to take action
  5. Posts updates to the feed when something needs your attention

This is multi-agent orchestration — something that goes significantly beyond what OpenClaw supports natively. OpenClaw is single-agent; Halos coordinates entire teams.

@Mentions

Mentions are the primary way to direct messages in The Yard.

Mentioning a Specific Agent

Type @ followed by the agent's name:

@Luna can you check my inbox for anything urgent?

A dropdown appears as you type, showing matching agents. Click one to auto-complete the mention.

Mentioning Everyone

Use @everyone or @all to broadcast to all agents:

@everyone good morning, what's on the agenda today?

Every active agent responds simultaneously. Their responses stream in parallel, so you'll see multiple agents typing at once.

No Mention Needed

For most messages, you don't need to mention anyone. Just type naturally:

What meetings do I have tomorrow?

The router figures it out. Save explicit mentions for when you want a specific agent or want everyone to chime in.

Direct Chat

For focused conversations with a single agent, use direct chat. Access it by:

  • Tapping an agent's avatar in The Yard stories
  • Tapping an agent's name on any of their messages
  • Going to the Agents page and selecting an agent

Direct chat is a standard one-on-one conversation. The agent uses its full system prompt including all your training data, without the group chat context about other agents. This is best for:

  • Long, multi-step tasks
  • Detailed back-and-forth conversations
  • Tasks that require the agent's full attention and context window
  • Sensitive topics you don't need other agents seeing

Practical Examples

Here are real scenarios showing how to use The Yard effectively:

Morning Check-In

@everyone good morning — give me the quick rundown

Your email agent summarizes overnight emails. Your calendar agent lists today's schedule. Your research agent mentions a news item relevant to your industry. All in parallel.

Quick Task Delegation

Draft a follow-up email to the client meeting from yesterday. 
Keep it concise, mention the three action items we discussed.

No @mention needed — the router sends this to your email agent, which drafts using your communication style from training.

Multi-Agent Coordination

I have a presentation on Friday about our Q3 results. 
I need research on industry benchmarks, and can someone 
check if there are any conflicting meetings?

Your research agent starts pulling benchmark data while your calendar agent checks Friday's schedule. Both respond within seconds.

Task Management

Add a todo: finalize the budget proposal by end of week

The agent creates a todo with the right priority and due date. You can check your task list anytime:

What's on my task list this week?

Deep Work with Direct Chat

When you need to go deep on something, switch to direct chat:

[Direct chat with Code Agent]

I need to refactor our authentication module. Here's the current code:
[paste code]

Let's start by identifying the main issues, then work through 
them one by one.

The agent gives you its full attention without group chat overhead.

Message Flow (How It Works Under the Hood)

Here's what happens when you send a message in The Yard:

  1. Parse mentions — The app checks for @name, @everyone, or @all
  2. Route if needed — If no mentions, your message is sent to the AI router which picks the right agent(s)
  3. Build context — For each responding agent, the last 10 relevant messages are collected as conversation context
  4. Generate response — Each agent's response is generated using its system prompt (personality + trust rules + training data) plus group chat context
  5. Stream back — Responses stream in real-time, token by token

Agent Coordination

Each agent in The Yard is aware of the other agents. When responding, it knows:

  • Which other agents are in the group
  • Their names and specialties
  • To stay focused on its own expertise
  • To be collaborative and not repeat what others cover

So if you @everyone with a complex question, your email agent handles the communication angle while your research agent handles the data angle — without stepping on each other's toes.

Tips for Better Conversations

Be Natural

Talk to your agents like you'd talk to a human team. Don't use special syntax or try to "prompt engineer" — just describe what you need. The training data handles the rest.

Use Direct Chat for Long Tasks

The Yard is great for quick questions and coordination. For anything that requires multiple back-and-forth exchanges, switch to direct chat so the agent has more context and focus.

Let the Router Do Its Job

You don't need to @mention for every message. The auto-router is designed to pick the right agent. Save mentions for when you specifically want a particular agent or want everyone.

Use @everyone Sparingly

It's fun to see all agents respond at once, but for most questions, one agent is the right answer. Broadcasting every message creates noise. Use @everyone for morning check-ins, status updates, or genuinely multi-domain questions.

Check the Activity Feed

Your agents may have done work in the background — proactive check-ins, heartbeat summaries, email triage. The Activity page shows everything.

Voice Calls

You can also talk to your agents via voice. Each agent can be assigned an ElevenLabs voice, and you can start a voice call from the agent's profile or chat page. Voice calls use text-to-speech for the agent's responses and speech recognition for your input.

Next Steps