OpenClaw custom provider setup is now managed in Lobsterland

Lobsterland can now configure a custom model endpoint for an OpenClaw instance without a manual openclaw.json edit. Choose the request adapter, enter the endpoint and model ID, declare the model's reasoning levels, and set its context window and maximum output tokens in the dashboard.

What does Lobsterland manage for an OpenClaw custom provider?

• Request adapter: OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages.
• Connection: base URL, API key, and model ID.
• Token limits: context window and output cap.
• Reasoning: whether the model supports configurable reasoning and the exact thinking levels it accepts.
• Lifecycle: configure at launch, add after launch, preserve imported values, or edit an existing profile in place.

How do you connect a custom provider to OpenClaw?

1. Choose Custom provider in the creation wizard, or open AI Model → Add provider on a running OpenClaw instance.
2. Choose the API adapter that the endpoint actually implements.
3. Enter the base URL, API key, model ID, context window, and output cap.
4. Enable configurable reasoning only when the endpoint documents it, select its supported thinking levels, then add and activate the provider.

Both token limits must be positive, and maximum output tokens cannot exceed the context window. Lobsterland checks those rules before saving the profile.

Which API adapter should you choose?

Compatibility is a contract, not a brand label. An endpoint described as “OpenAI-compatible” may implement Chat Completions without implementing Responses. Choose the route its documentation actually names.

How do context window and output cap differ?

The context window is the model's total working budget for the prompt, conversation, tool results, and generated answer. The output cap limits only the generated answer. Lobsterland defaults an unknown custom model to 200,000 and 8,192 respectively, but those defaults are not a compatibility promise. Replace them with the limits published by your endpoint.

How does custom-provider reasoning work?

When reasoning is enabled, Lobsterland writes the capability and selected effort levels into OpenClaw's model catalog. That lets OpenClaw validate /think choices against the configured model before sending a request. It does not add reasoning support to an endpoint that lacks it. Select only the levels the provider documents: Minimal, Low, Medium, High, Extra high, or Max.

Can you edit an existing custom provider?

Yes. Open AI Model, expand Configured providers, and choose Edit on the custom profile. You can change the adapter, base URL, model ID, token limits, and reasoning declaration without deleting the profile. Imported custom-provider metadata is also preserved when the source configuration contains it.

Where does this fit in a managed OpenClaw stack?

The provider connection is one part of the runtime. Lobsterland's managed OpenClaw hosting operates the always-on instance around it. If you need a broad hosted model catalog instead of one endpoint you control, follow the OpenRouter setup guide. The OpenClaw runtime page shows the wider managed surface.

OpenClaw custom provider setup FAQ

Which adapter should I choose?

Match the endpoint's documented route: Chat Completions for /v1/chat/completions, Responses for /v1/responses, or Anthropic Messages for /v1/messages.

How do context window and output cap differ?

The context window covers input and output together. Maximum output tokens caps only the generated answer and cannot exceed the context window.

Can I edit an existing provider without recreating it?

Yes. Use AI Model → Configured providers → Edit.

Does enabling reasoning make every thinking level work?

No. It declares capabilities the endpoint already has; it cannot create them.