Desk Note: Cursor Custom Provider Setup

The Cursor connection has three moving parts

Connecting Cursor to a custom provider means choosing a different service to handle supported model requests. Cursor supports gateways through a custom OpenAI-compatible base URL, but available models and features depend on the gateway and Cursor’s integration. Start with a simple chat test rather than assuming every editor feature will follow the new route. ([help.openai.com](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor))

Keep three pieces of information together: the provider’s API key, its base URL, and the exact model identifier it accepts. Think of these as your access pass, destination, and requested service. A provider’s familiar display name is not necessarily the identifier you must enter in Cursor. The gateway setup requires credentials, an endpoint, and a corresponding model or deployment name. ([help.openai.com](https://help.openai.com/en/articles/20001506-using-openai-models-in-cursor))

Prepare Relay365’s endpoint without guessing model IDs

Relay365 is a pay-as-you-go relay desk covering model families including Claude, GPT, Gemini, DeepSeek, Kimi, GLM, Qwen, and Grok. For the Relay365 configuration supplied here, the API base is https://api.api365relay.com, and OpenAI-compatible clients use https://api.api365relay.com/v1.

Before changing Cursor, obtain your Relay365 key and copy an available model’s exact API identifier from the provider’s account interface or documentation. If you want to try Gemini 2.5 Flash, confirm the accepted identifier and its compatibility with Cursor first. Do not invent an identifier by lowercasing the display name or adding a provider prefix.

Set Cursor’s key, base URL, and model together

The manual custom-provider flow uses Cursor’s OpenAI API key controls and base URL override. Record your existing settings before replacing them so you have a clear rollback path. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))

• Open Cursor Settings, select Models, and find the API Keys controls. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))
• Enter your Relay365 key in the OpenAI API key field and enable custom-key use. Although the field says OpenAI, this setup pairs it with the custom endpoint. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))
• Enable Override OpenAI Base URL and enter https://api.api365relay.com/v1 as the destination.
• Add the exact provider-issued model identifier to Cursor’s model list, or enable the matching entry if it already exists. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))
• Save the configuration, select that model, and start a new standard chat for testing. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))

Confirm one small request before using project context

Use a harmless prompt such as “Reply with one short sentence.” After the response, inspect the provider’s usage records to confirm the request reached the intended account. Starting a new chat after an endpoint or model change helps avoid testing an old selection. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))

If the test fails, check the key, endpoint, and model identifier separately. Confirm that the base URL includes /v1, and test the same credentials and model with the provider’s documented request example. This separates a provider-access problem from a Cursor-specific configuration problem. Avoid changing several settings simultaneously. ([fast.inference.net](https://fast.inference.net/docs/integrations/cursor))

Keep Cursor’s feature and privacy boundaries visible

A custom chat connection does not reroute all of Cursor. Cursor documents that custom keys work with chat models, while Tab completion continues using its built-in models. Provider model charges are separate, and additional Cursor charges can depend on your plan. ([prod.cursor.com](https://prod.cursor.com/help/models-and-usage/api-keys))

Review data handling before sending private code. Cursor says requests using your key still pass through its backend for prompt building, and its Zero Data Retention policy does not apply to bring-your-own-key use. Keep credentials out of project files, prompts, and shared screenshots. ([prod.cursor.com](https://prod.cursor.com/help/models-and-usage/api-keys))

Leave the next shift a repeatable configuration

Relay365 fits when you want pay-as-you-go access across several model families. If that matches your workflow, create a Relay365 account and validate one supported model before expanding usage. Keep a short setup note containing the endpoint, exact model identifier, test result, and rollback steps—but never the key itself.