HalosHalosDocs

Deploy an Agent

Deploying a new agent takes about 30 seconds. You can start from scratch by picking a specialty, or install a pre-configured playbook. Either way, your agent is live in the cloud almost instantly.

Two Ways to Deploy

1. From Scratch

Pick a specialty, customize the personality, and deploy. Best when you want full control from the start.

2. From a Playbook

Install a pre-configured agent bundle that comes with skills, permissions, scheduled tasks, and configuration already set up. Best for common use cases where someone has already figured out the optimal setup. See Playbooks for details.

Deploying from Scratch

Step 1: Choose a Specialty

When you click Deploy New Agent (the + button in The Yard, or the Agents page), you'll first pick what your agent specializes in:

| Specialty | Best For | Example Tasks | |-----------|----------|---------------| | Email | Inbox management | Triage, draft replies, organize, apply labels, flag urgent items | | Calendar | Schedule management | Manage appointments, detect conflicts, send reminders, prep meeting briefs | | Research | Information gathering | Search the web, summarize articles, track trends, compile reports | | Code | Software development | Code review, debugging, generation, explain codebases, PR summaries | | Photo | Photo library | Organize, tag, detect duplicates, create albums | | Custom | Anything else | General-purpose — adapts to whatever you train it on |

The specialty determines the agent's default system prompt, starting skills, and initial permission set. You can always customize these later through training. See Agent Types for a deep dive on each one.

Which Specialty Should You Pick?

Pick based on your biggest pain point. If you spend 2 hours a day on email, an email agent will give you the most immediate value. If you're constantly Googling things mid-task, research is your best bet.

Custom agents are powerful but need more training. A specialty agent works well out of the box because it has domain-specific defaults. A custom agent starts more generic — it shines once you've trained it with knowledge and instructions specific to your use case.

Step 2: AI-Assisted Creation

Halos uses AI to help you configure your agent. When creating a new agent, the AI assistant can:

  • Suggest a name based on the agent type (from a curated pool of distinctive names like Luna, Atlas, Nova, Sol)
  • Generate personality settings based on how you describe what you want
  • Recommend skills based on your use case
  • Suggest integrations that would work well with the agent type

You can also describe what you want in natural language, and the platform will suggest the best agent type, configuration, and skills.

Step 3: Name & Appearance

Give your agent a name and pick an avatar (emoji) and accent color. These are cosmetic but they matter more than you'd think — they make it easy to identify agents at a glance in The Yard.

Naming tips:

  • Use distinctive names — you'll type these with @mentions in chat, so keep them short
  • Names like "Luna" or "Atlas" are easier to type than "EmailAgent01"
  • The name suggestions refresh each time if you want inspiration
  • You can pick any emoji as an avatar, not just the presets

Step 4: Connect Integrations (Optional)

Some agent types work best with access to your real data. For example, an email agent can connect to your Gmail via OAuth to actually read and organize your inbox.

Integrations are optional — your agent works without them, it'll just rely on what you tell it in chat and what you provide via training data.

Currently supported:

| Integration | What It Enables | |------------|-----------------| | Gmail | Read, organize, label, and (at higher trust) send emails | | Google Calendar | View and manage your schedule | | Apple Calendar | View events and create new ones | | Calendar URL | Subscribe to any ICS/webcal calendar feed | | Notion | Access your Notion workspace, databases, and pages | | Slack | Read and interact with your Slack channels | | GitHub | Access repositories, PRs, and issues |

You can connect integrations during deployment or later from Settings > Connections. See Integrations for setup guides.

Step 5: Deploy

When you hit deploy, here's what happens:

  1. Database record — Your agent's configuration is stored
  2. Skills installed — Default skills are loaded based on the agent's specialty
  3. Permissions set — Starting at the Observer trust tier (read-only)
  4. Knowledge access — A dedicated memory collection is created for the agent
  5. API access granted — Available API tools are configured
  6. Ready to chat — The agent appears in your Yard

The entire process takes a few seconds. You'll see a progress animation while it deploys.

After Deployment

Once your agent is live, here's what you can do:

Chat

Start a conversation right away — either directly at the agent's chat page or in The Yard. Try asking it what it can help with:

Hey [agent name], what are you good at?

Train

Head to the agent's profile page and add training data. Even 2-3 pieces of knowledge and one instruction make a noticeable difference. See Training for details.

Configure

On the agent's profile page you can adjust:

  • Personality — Tone, verbosity, and communication style
  • Model — Which AI model powers the agent (Claude, GPT-4o, Gemini, and more)
  • Voice — Choose an ElevenLabs voice for voice calls
  • Heartbeat — How often the agent proactively checks in (disabled, 15 min, 1 hour, 8 hours, daily)
  • Email address — Give the agent its own email address for receiving external mail
  • Telegram — Connect a Telegram bot for mobile chat
  • Objective — Set the agent's primary mission
  • Background — Give the agent a backstory and context

Install Skills

Browse the Skills library and add capabilities. Skills are hot-installed — no restart needed. See Skills & Trust.

Sleep & Wake

If you don't need an agent right now, you can sleep it to save resources. Sleeping an agent pauses it but keeps all configuration and training data. Wake it when you need it again — it picks up right where it left off.

Managing Multiple Agents

There's no hard limit on how many agents you can deploy. Each one is fully independent with its own:

  • Training data and knowledge base access
  • Installed skills and permissions
  • Personality and configuration
  • Usage stats and XP tracking
  • Connected integrations
  • Email address and Telegram bot
  • Memory collection

The Agent Team

The power of multiple agents shows up in The Yard. When you have an email agent, a calendar agent, and a research agent all active, you can type natural messages and the platform routes them intelligently:

What meetings do I have tomorrow and are there any emails I need to handle before them?

This message might trigger both your calendar agent (for the meetings) and your email agent (for relevant emails) to respond in parallel. Meanwhile, your head agent coordinates the team behind the scenes.

Agent Profiles

Every agent has a profile page showing:

  • Stats — Tasks completed, time saved, accuracy
  • Level & XP — Progress toward the next level
  • Training — All knowledge, instructions, and examples
  • Skills — Installed capabilities with individual skill levels
  • Permissions — What the agent can and can't do
  • Memory — What the agent remembers from conversations
  • Activity — Recent actions and interactions
  • Configuration — Model, personality, heartbeat, email, Telegram settings

Deleting an Agent

If you no longer need an agent, you can delete it from its profile page. This:

  • Removes all training data, memories, and skills
  • Removes all agent-specific knowledge access
  • Is permanent and cannot be undone

If you just want to pause an agent temporarily, use Sleep instead.

Common Questions

How long does deployment take? Usually under 30 seconds. You'll see a progress animation.

Can I rename an agent after deployment? Yes — go to the agent's profile page and update the name, avatar, or color anytime.

Do agents consume resources when I'm not chatting? Active agents use minimal resources when idle. Heartbeat check-ins use a small amount of LLM tokens at each interval. If you want to fully pause an agent, use Sleep mode.

Can agents talk to each other? In The Yard, agents are aware of each other and coordinate their responses. The head agent actively manages the team. They won't duplicate effort on a question that another agent is better suited to answer.

Can I export my agent's setup? Yes — you can export any agent as a Playbook, which captures its configuration, skills, permissions, and scheduled tasks for reuse or sharing.

Next Steps

Your agent is deployed — now make it smarter.