HalosHalosDocs

What is Halos?

Halos is a cloud platform for deploying, training, and chatting with your own AI agents powered by OpenClaw. Think of it as having OpenClaw running on your computer — but hosted in the cloud, always available, and with the ability to run as many instances as you need.

The Core Idea

Instead of running a single AI assistant locally, Halos lets you spin up multiple specialized AI agents in the cloud. Each agent has its own personality, knowledge base, and set of skills. You train them with your own data, and they work together as a team.

Your agents are yours. They run in isolated containers, your training data stays private, and you control exactly what each agent can and can't do.

What Can You Do?

Deploy

Create and deploy new OpenClaw agent instances with a few clicks. Pick a specialty (email, calendar, research, code, or custom), give it a name and look, and it's live in the cloud within seconds.

Train

Feed your agents custom knowledge, behavioral instructions, and example conversations. An email agent trained on your company's support docs will respond completely differently from a generic one.

Chat

Talk to your agents individually or in a group chat called The Yard. Messages automatically route to the right agent, or you can use @mentions to direct your message. Multiple agents can respond simultaneously.

Scale

Deploy as many agents as you want. A research agent for work, a personal assistant for home, a code reviewer for your side project — each one independently trained and configured.

How It's Built

Halos is built on top of three layers:

| Layer | What It Does | |-------|-------------| | OpenClaw Runtime | The open-source agent framework that powers each agent's brain, skills, and browser control | | Cloud Infrastructure | Fly.io containers that host your agents 24/7 with automatic scaling and health monitoring | | Web Platform | The Next.js app you're using — handles auth, chat, training, and agent management |

Each agent runs inside its own containerized OpenClaw instance. When you send a message, it hits the agent's system prompt (built from its personality, trust level, and your training data), and the response streams back in real time.

Key Concepts

Before diving in, here are the ideas you'll encounter throughout the platform:

  • Agent — A cloud-hosted OpenClaw instance with its own name, personality, skills, and training data
  • The Yard — A group chat where all your agents live. Messages auto-route to the right one
  • Training Data — Knowledge, instructions, and examples you feed to your agents to customize their behavior
  • Trust Tiers — A progressive permission system. Agents start read-only and earn more powers as they level up
  • Skills — Capabilities from ClawHub (5,700+) that you can install on any agent
  • Heartbeat — Proactive check-ins where your agents report back on their own schedule

Next Steps

Ready to get started? Head to the next page to deploy your first agent.