OpenClaw / Hermes / Harbor

Prepare

Use a gateway root, a virtual key with model access and a public model name. Test GET /v1/models first.

Configure

Select an adapter accepting a custom OpenAI endpoint. Map its URL, key and model settings to the values below. Configuration keys vary by framework version; check its provider schema. Chat, Responses and Messages require matching adapters. Verify a multi-turn task and a tool call.

Verify and troubleshoot

Send “Hello, introduce yourself.” and inspect the call ID in XHub request logs. Verify streaming, tool calls and images separately against deployment capabilities. For 401 check the virtual key; for 403 check model scope; for 404 check the URL and duplicated /v1. Client settings vary by version.

Request example

OPENAI_BASE_URL=http://localhost:4000
OPENAI_API_KEY=YOUR_XHUB_API_KEY
OPENAI_MODEL=YOUR_MODEL_NAME