Cursor

Prerequisites

• Install the published @unlimgpt/cli package.
• Install Cursor and confirm it offers an OpenAI-compatible endpoint and API key field.

Setup

• Run unlimgpt configure cursor and open Settings → Models → API Keys.
• Test Cursor Local IDE in Ask first: Agent with a custom OpenAI endpoint is not guaranteed; Cloud Agents do not use BYOK keys.
• Enable Use OpenAI API Key and Override OpenAI Base URL; enter the exact model ID without a prefix.
• To choose effort from Cursor's model picker, append -effort-low and -effort-high to the model ID shown by the CLI through Add Custom Model. For example: gpt-6-sol-effort-low and gpt-6-sol-effort-high.
• Cursor's separate Reasoning control may omit reasoning_effort on a custom Base URL; choose the corresponding model entry.
• Copy the displayed Base URL and model ID. The CLI temporarily copies the API key to the clipboard.
• Paste the API key from the clipboard only into Cursor's native secret field; the CLI never reads SQLite or VS Code SecretStorage.

Verify

• Open Cursor, send “Reply with READY” through the configured model, wait for a non-empty response, and confirm the request in API Activity. The wording may vary; the request uses credits.

Rotate

• Record the old key ID from unlimgpt keys list, create a new key with unlimgpt keys create --name cursor-next, then rerun unlimgpt configure cursor to copy the new key.
• Paste it into Cursor's native secret field. Send a real request in Cursor and confirm it in API Activity; only then revoke the old key on the API keys page.

Revoke

• Revoke a key on the API keys page after confirming the new connection; the old key immediately loses access.

Revert

• The CLI did not modify Cursor; remove the key and model through its UI.

Troubleshooting

If the request fails, run unlimgpt doctor --client cursor and see troubleshooting. Diagnostics do not replace a real request.

Uninstall

• Remove the UnlimGPT profile and secret through Cursor's UI, then revoke the dedicated API key.