Overview
OpenClaw is a personal AI runtime with an OpenAI-compatible API and a tool registry. The OpenClaw pack exposes three actions for agents to connect to and orchestrate a user’s OpenClaw instance. Use cases- Calling personal OpenClaw-hosted models from within Agent.ai workflows
- Invoking custom tools you’ve registered on your OpenClaw instance
- Bridging enterprise private models with Agent.ai’s orchestration layer
Pack actions
| Action | Purpose |
|---|---|
openclaw_get_details | Fetch the user’s OpenClaw instance IP, auth token, and status. Reference downstream as {{ openclaw.instance_ip }} / {{ openclaw.auth_token }}. |
openclaw_chat_completion | Send a chat completion request (OpenAI-compatible). Returns the assistant’s response text. |
openclaw_tools_invoke | Invoke a single registered tool on the instance via the Tools Invoke API. |
Typical flow
Tips
- Always call
openclaw_get_detailsfirst so downstream actions have the instance metadata cached as variables. - Handle the
statusfield — skip subsequent calls if the instance is offline.
Related
- Full endpoint details in the API Reference.