Cursor

Cursor can send OpenAI-compatible requests through two endpoints:

• Standard chat models use /v1/chat/completions.
• GPT-5.x reasoning models in Agent use /v1/responses.

Configure the API root, https://api.tryarcflow.com/v1, so Cursor can use the correct endpoint for the selected model.

1. Configure the OpenAI provider

Open Cursor Settings → Models (Cmd+Shift+J on macOS), then:

1. Enable OpenAI API Key and enter your Arcflow sk-... key.
2. Enable Override OpenAI Base URL.
3. Enter exactly https://api.tryarcflow.com/v1.
4. Click Verify or Save.

Do not append /chat/completions or /responses to the base URL.

If requests fail to connect through a custom gateway, open Cursor Settings → Network and try HTTP Compatibility Mode → HTTP/1.1.

2. Choose a model

Select a built-in GPT model whose ID is available from Arcflow, such as:

• gpt-5.5
• gpt-5.4
• gpt-5.6-sol
• gpt-5.6-terra
• gpt-5.6-luna

Do not add these IDs as custom models. Cursor rejects a custom ID when it duplicates, or contains the name of, one of its built-in models.

Cursor's Refresh Model List refreshes Cursor's catalog; it does not import models from Arcflow's /v1/models endpoint. The built-in model ID is sent to Arcflow when you select it.

Provider limitations

The OpenAI key and base URL override apply globally to OpenAI-family models; Cursor does not provide a per-model API-key selector.

• Grok, Composer, and Auto are Cursor-hosted models and do not support the custom OpenAI key. Disable the OpenAI key (Cmd+Shift+0) before selecting one of them.
• Built-in Claude models use Cursor's Anthropic provider, not the OpenAI base URL override. An Arcflow Claude model can use the OpenAI-compatible route only when its model ID can be added as a non-colliding custom model.

If Cursor reports Unsupported parameter: user on a GPT-5.x Agent request, contact Arcflow support. Current Arcflow deployments remove Cursor's tracking field before forwarding Responses API requests.