n8n

n8n can call Modelship two ways: through its built-in OpenAI-shaped nodes, or directly via HTTP.

Option 1 — OpenAI Chat Model node

n8n's OpenAI Chat Model node (used by its AI Agent / LangChain nodes) takes an OpenAI credential with a configurable Base URL:

1. Create a new credential of type OpenAI API.
2. Set Base URL to http://<modelship-host>:8000/v1.
3. Set API Key to a real key if MSHIP_API_KEYS is configured, or any non-empty placeholder if not.
4. Use that credential in the OpenAI Chat Model node and set Model to a name from your models.yaml.

This is the fastest path if you're building with n8n's AI Agent nodes and want tool calling handled by n8n's own agent loop.

Option 2 — HTTP Request node

For direct control over the request (e.g. to call /v1/responses for server-side conversation state, which n8n's built-in node doesn't expose), use the HTTP Request node:

• Method: POST
• URL: http://<modelship-host>:8000/v1/responses
• Authentication: Generic → Header Auth, header Authorization = Bearer <key> (if MSHIP_API_KEYS is set)
• Body (JSON):

{
  "model": "reasoning-qwen",
  "input": "{{ $json.userMessage }}",
  "previous_response_id": "{{ $json.previousResponseId }}"
}

Feed resp.id from the response back into the next call's previous_response_id to keep a conversation going across workflow runs without re-sending history.