Use inference snaps in OpenClaw

OpenClaw is a local AI agent that supports tool calling. This guide will help you configure OpenClaw to use an inference snap.

Similar steps can be used to configure other AI agents, such as PicoClaw, NemoClaw, and Hermes.

You must have OpenClaw and an inference snap installed to continue.

Get the inference snap endpoint and model name

Use the inference snap’s status command to retrieve the OpenAI endpoint URL and model name.

For example, using the qwen3 snap:

qwen3 status

The output will show the endpoint and model information:

engine: cpu
services:
    server: active
    server-webui: active
endpoints:
    openai: http://127.0.0.1:8338/v1
    webui: http://127.0.0.1:8339/
model:
    name: qwen3-8b-q4-k-m

Note down the openai endpoint URL and the model.name value, as you’ll need them for configuration.

Configure OpenClaw

After installation, OpenClaw launches the interactive setup wizard with a number of sections to set up the agent. The connection with an inference snap needs to be set up during the model configuration section.

If OpenClaw has already been set up, the wizard can be manually started again to run through the model configuration section:

openclaw configure --section model

The wizard will ask a number of questions. Answer the questions using this guide as a reference:

  • Model/auth provider: Under the “More…” section, choose “Custom Provider”.

  • API Base URL: Enter the openai endpoint URL obtained previously from the status command.

  • How do you want to provide this API key?: Choose “Paste API key now”

  • API Key: Leave blank

  • Endpoint compatibility: Choose “OpenAI-compatible”.

  • Model ID: Enter the model.name value obtained previously from the status command.

The wizard will now run a live completion probe to verify the endpoint. If it fails, it could be due to insufficient RAM or VRAM, or missing tool-calling support in the model. Refer to the OpenClaw troubleshooting guide for further help.

  • Endpoint ID: Arbitrary identifier of the provider, e.g. set it to qwen3-snap.

  • Model alias (optional): An alias for the model. It falls back to the model ID if left unset.

The wizard should print Configure complete. You can now use OpenClaw, and it should use the inference snap as its model.

If you have multiple models configured, the OpenClaw dashboard lets you choose the one to use from the dropdown below the chat input box.