Skip to main content
OpenClaw is an open-source AI assistant gateway. It runs as a single always-on process that connects AI models to messaging platforms, routing conversations through configurable agents with specialized skills.

​
Gateway architecture

OpenClaw runs as a persistent process with a WebSocket control plane on port 18789. The gateway:
  • Maintains long-lived connections to messaging platforms (Discord, Telegram, Slack).
  • Receives incoming messages and routes them to the appropriate agent.
  • Sends inference requests to the configured AI provider.
  • Returns responses through the originating channel.
On Operator, each instance runs one OpenClaw gateway process in an isolated Azure container. The control plane is used internally for config sync and health checks — you interact with your instance through the Operator dashboard, not the WebSocket directly.

​
Multi-channel support

OpenClaw connects to multiple messaging platforms simultaneously from a single instance. On Operator, the supported channels are:
  • Discord — Bot accounts via the Discord Gateway API. Users DM the bot directly.
  • Telegram — Bot accounts via the Telegram Bot API. Users chat with the bot in private messages.
  • Slack — App accounts via Socket Mode. Users DM the bot in their workspace.
Each channel has its own authentication token, allow list, and pairing policy. See Channels for setup instructions.

​
Multi-agent routing

A single OpenClaw instance can run multiple agents. Each agent has its own:
  • Persona — System prompt, name, and behavior configuration.
  • Channel bindings — Which channels the agent responds on.
  • Skills — Which tools and capabilities the agent can use.
  • Context — Conversation history and memory.
You can route different channels to different agents, or have multiple agents handle different types of requests on the same channel. Agent configuration is managed through the Operator dashboard.

​
Skills system

Skills extend what your assistant can do beyond conversation. OpenClaw has four skill sources:
  • Bundled skills — Built into OpenClaw. Basic capabilities like conversation memory, context management, and message formatting.
  • Managed skills — Maintained by the OpenClaw project. Installable via the dashboard (e.g., web search, code execution, file handling).
  • Workspace skills — Custom skills you define in your instance’s workspace. Stored in the instance’s Azure Files volume.
  • ClawHub registry — Community-contributed skills you can browse and install.
Skills are enabled or disabled per agent, so different agents in the same instance can have different capabilities.

​
Configuration

OpenClaw instances are configured via openclaw.json. This file defines:
  • Agents and their personas
  • Channel connections and tokens
  • Enabled skills per agent
  • Provider settings (auto-configured on Operator)
On Operator, you do not edit openclaw.json directly. The dashboard generates and syncs the configuration to your instance’s Azure Files volume. When configuration changes, the container restarts automatically to pick up the new settings.

​
How Operator runs OpenClaw

When you create an instance on Operator:
  1. Operator provisions an Azure Container App with the latest OpenClaw image.
  2. An API key and operator provider configuration are injected into the container.
  3. The operator provider points to Spider for AI inference — no external API keys needed.
  4. Configuration from the dashboard is written to Azure Files and mounted into the container.
  5. The OpenClaw process starts and connects to your configured channels.
You manage everything through the dashboard. Operator handles container lifecycle, config sync, version upgrades, and infrastructure.

​
Next steps

  • Instances — Instance lifecycle, limits, and configuration management.
  • Channels — Connect Discord, Telegram, and Slack.
  • Security — How your data is encrypted and isolated.