Cursor

Cursor lets you supply your own OpenAI-compatible endpoint. Once configured, Cursor's chat calls LuluTokens instead of Cursor's own backend.

This replaces Cursor's chat models only. Tab completion and some agent features run on Cursor's proprietary models and are unaffected by a custom key — they keep using your Cursor plan.

Configure

Open the model settings

Cursor Settings → Models (Cmd+Shift+J on macOS, Ctrl+Shift+J elsewhere).

Add the model IDs first

Scroll to the model list and add each LuluTokens model ID you want, exactly as it appears in GET /v1/models — for example deepseek-v4-pro.

Do this before entering the key: Cursor verifies a custom endpoint by sending a real request using a model from this list, and verification fails if the list only contains OpenAI model names the gateway does not serve.

Enter the key and base URL

In the OpenAI API Key section:

Field	Value
API Key	your `sk-...` key
Override OpenAI Base URL	`https://api.lulutokens.ai/v1`

Then press Verify. A green check means Cursor reached the gateway.

Select the model in chat

Open the chat panel and pick your model from the dropdown. Disable the models you do not intend to use, so Cursor does not silently fall back to one of them.

Notes and limits

Verification is strict. Cursor's check fails on any non-200, including a 402 from an empty balance. Confirm the key works with curl first.
Long context is capped by Cursor, not by us — Cursor trims the prompt it sends regardless of the model's real context window.
Privacy mode still applies to Cursor's own features. Requests that go to LuluTokens are covered by our terms; the rest are covered by Cursor's.

Troubleshooting

Symptom	Fix
Verify button fails	A model in the list is not served by the gateway — remove it
`404` after verifying	Base URL lost its `/v1` suffix when saved
Chat ignores your model	Disable the built-in models so Cursor cannot fall back
Tab completion unchanged	Expected — Tab does not use custom endpoints