Cursor
Cursor lets you override the OpenAI base URL, so its chat can talk to OpusGate's OpenAI-compatible side. Configuration lives inside Cursor's own settings UI (there is no config file to script — set it up once by hand).
Required values
Base URL:
API key: your
https://api.opusgate.dev/v1 — keep the /v1, Cursor appends /chat/completions itself.API key: your
sk-og-... from the Keys page.Setup
- Open Settings (
Cmd/Ctrl + ,) → Models in the left sidebar. - Click + Add Model and type the exact model ID you want (e.g.
claude-opus-4.8,claude-sonnet-4.6). IDs are case-sensitive — copy them from the models page. - Turn on Override OpenAI Base URL and enter
https://api.opusgate.dev/v1. - In the OpenAI API Key field, paste your
sk-og-...key. (Despite the "OpenAI" label, the key goes to OpusGate.) - Click Verify — Cursor sends a test request through OpusGate. A green check means you're done.
What routes through OpusGate — and what doesn't
- Only the Chat / Plan panel (
Cmd/Ctrl + L) uses the custom endpoint. This is a deliberate Cursor limitation, not an OpusGate one. - Agent, Edit/Composer, Apply and autocomplete stay on Cursor's backend and keep billing to your Cursor subscription.
- Want full agentic coding at OpusGate prices? Use Claude Code, Cline or Roo Code — they route everything through your key.
Troubleshooting
Verify fails
Three usual causes: the base URL is missing /v1; the key was pasted with a space or is revoked (issue a fresh one); or your balance is empty (top up — verification makes a real, tiny billed request).
Model responds with "model not found"
The model name you added must match an OpusGate ID exactly — claude-opus-4.8, not claude-3-opus or a display name. Check the catalog.
Custom model works, then Cursor switches back
Cursor re-enables its built-in models after some updates. Re-select your custom model in the picker at the bottom of the chat panel.