Skip to content
BrainRoad BrainRoad
Documentation Menu

What Is OpenClaw?

OpenClaw is the open-source AI agent runtime that powers every BrainRoad agent.

On this page

The Agent Runtime Behind BrainRoad

Every agent you create on BrainRoad is powered by OpenClaw — an open-source AI agent framework with over 100,000 GitHub stars, released under the MIT license. Created by Peter Steinberger (now at OpenAI), OpenClaw describes itself as “Your own personal AI assistant. Any OS. Any Platform.”

OpenClaw handles the AI work: processing conversations, running tools, browsing the web, and connecting to messaging channels like WhatsApp, Signal, iMessage, and email. BrainRoad handles everything around it — infrastructure, security, lifecycle management, and the web dashboard you use to control your agent.

Gateway Architecture

At the core of OpenClaw is a single gateway process that runs on port 18789. This gateway acts as a WebSocket control plane, connecting your messaging channels to the AI agent runtime (called Pi).

The data flow looks like this:

Channels → Gateway → Agent (Pi)

Messages arrive from any connected channel (WhatsApp, Signal, the Chat tab, etc.), pass through the gateway, and reach the agent runtime. The agent processes the message using your configured AI model and sends a response back through the same path.

Configuration lives in two files inside your agent’s container:

  • ~/.openclaw/openclaw.json — main configuration (JSON5 format, hot-reloadable). Controls model selection, personality, channels, skills, and behavior.
  • ~/.openclaw/agents/main/agent/auth-profiles.json — API key storage. Holds credentials for AI providers like Anthropic, OpenAI, and Google.

What BrainRoad Adds

OpenClaw is designed to run on your own machine. BrainRoad turns it into a managed cloud service:

  • Isolated Kubernetes containers — each agent runs in its own namespace with dedicated resources. No shared state, no noisy neighbors.
  • Persistent storage — your agent’s files, conversation history, and configuration survive restarts, updates, stop/start cycles, and even plan changes. Nothing is lost.
  • Web dashboard — manage your agent from a browser. No terminal, no SSH, no DevOps experience required.
  • Automatic updates — when OpenClaw releases improvements, BrainRoad updates the gateway image. Your agent benefits without you lifting a finger.
  • API key management — add and rotate keys through a visual interface instead of editing JSON files by hand.
  • Zero infrastructure — no servers to provision, no ports to open, no Docker to install, no processes to babysit.

Why Managed Hosting Matters

Self-hosting OpenClaw is straightforward on a technical level, but the security implications are serious. Over 42,000 exposed OpenClaw instances have been found online — many with plaintext API credentials accessible to anyone who knows where to look.

The risks of self-hosting include:

  • Plaintext credentials — API keys for Anthropic, OpenAI, and other providers are stored in a JSON file. If your instance is reachable from the internet, those keys are exposed.
  • Localhost trust assumptions — OpenClaw’s default configuration trusts connections from localhost. Running it on a VPS or cloud instance without proper firewalling exposes the control plane.
  • Exposed ports — the gateway listens on port 18789. Without a reverse proxy and authentication layer in front of it, anyone can connect to your agent.

BrainRoad eliminates these risks. Each agent runs behind Cloudflare Access authentication, inside an isolated Kubernetes namespace, with no ports exposed to the public internet. Your API keys never leave your agent’s container.

Supported AI Providers

OpenClaw works with every major AI provider. You can switch models at any time through the dashboard or via the /model chat command:

  • Anthropic — Claude Opus, Sonnet, Haiku (recommended)
  • OpenAI — GPT-4o, GPT-4, o1, o3
  • Google — Gemini Pro, Gemini Flash
  • OpenRouter — access to hundreds of models through a single API key
  • xAI — Grok models
  • Local models — Ollama and any OpenAI-compatible API endpoint

Skills Ecosystem

OpenClaw supports installable skills that extend what your agent can do. The ClawHub registry hosts over 50 community-built skills, including integrations for:

  • Productivity — 1Password, Notion, Obsidian, Trello, Todoist
  • Development — GitHub, GitLab, Linear, Jira
  • Communication — Slack, Discord, Telegram
  • Data — Google Sheets, Airtable, Postgres
  • Web — Browser automation, web scraping, RSS feeds

You can install skills through the Console tab or let the agent discover and install them itself using the find-skills tool.

Learn More