Skip to content
BrainRoad BrainRoad
Documentation Menu

How Agents Work

Your agent runs in an isolated container with its own filesystem, browser, and messaging connections.

On this page

One Agent, One Isolated Environment

When you create an agent on BrainRoad, the platform provisions a dedicated container in its own isolated namespace. Your agent does not share resources, storage, or configuration with any other user. It is a private environment that belongs entirely to you.

Under the hood, each agent runs OpenClaw — an open-source AI agent runtime. BrainRoad handles the infrastructure layer: provisioning, networking, persistent storage, lifecycle management, and the web dashboard. OpenClaw handles the AI work: processing conversations, browsing the web, running tools, and connecting to messaging channels.

What Lives Inside Your Agent

Each agent container includes:

  • Persistent filesystem — your agent’s files, conversation history, and configuration are stored on dedicated storage that survives restarts, updates, and even plan changes. Nothing is lost when your agent stops.
  • A browser instance — your agent has its own headless browser for web research, form filling, and any task that requires navigating websites. You can watch it work in real time via the Browser tab.
  • Messaging connections — your agent can connect to WhatsApp, Signal, iMessage, and email. Once connected, it receives and responds to messages independently.
  • Configuration files — your agent’s personality, model preferences, and behavior rules are stored in a config file inside the container. Changes to this file take effect immediately without restarting.
  • Pre-installed developer tools — your agent ships with Claude Code CLI, OpenAI Codex CLI, Homebrew, uv (Python), and Bun. These are ready to use from the Console with no installation required.

The Thin Wrapper Model

BrainRoad is intentionally a thin layer around OpenClaw. The platform collects your API keys, writes them into your agent’s container, manages start/stop lifecycle, and proxies the dashboard interface. It does not modify or fork the underlying agent runtime.

This means you get the full power of OpenClaw with zero infrastructure management. As OpenClaw improves, your agent benefits automatically when BrainRoad updates the gateway image.

For a deeper look at what OpenClaw is, its architecture, and the ecosystem of skills and channels it supports, see What Is OpenClaw?.

Resource Limits

Each agent operates within defined resource limits to keep the platform stable. These limits are generous enough for normal operation and scale with your plan tier.