Cursor

Cursor supports custom OpenAI-compatible endpoints for Pro subscribers. Register Nimbus once and every Nimbus model becomes selectable in the chat and inline-edit UIs.

Cursor's custom-model feature routes the request through your key against a base URL you specify — the wire format is OpenAI /v1/chat/completions, which Nimbus speaks natively. Function calling, streaming, and multimodal input all pass through.

Prerequisites

Cursor Pro subscription. The free tier does not expose custom endpoints — the "OpenAI API key" field is Pro-only.
Cursor 0.40+ (older builds shipped a narrower model list).
A Nimbus API key (sk-nim-...).

Install

Nothing to install — the built-in Cursor client handles the request. Grab Cursor from cursor.com if you don't have it yet.

Configure

Open Settings → Models.
Scroll to OpenAI API Key. Paste your Nimbus key (sk-nim-...) — Cursor treats any OpenAI-shaped endpoint the same way.
Toggle Override OpenAI Base URL and set:

bash

https://llm.nimbusapi.net/v1

Click Verify. A green check confirms Cursor reached Nimbus with the key.
Under Model Names, add each Nimbus model you want in the picker. Use the fully-qualified name — e.g. anthropic/claude-opus-4.8, openai/gpt-5.1.
Uncheck every non-Nimbus model to force Cursor to route everything through your key.

Short-alias table

Nimbus accepts short aliases in the model field so you don't have to type the provider prefix every time in Cursor's dropdown. The aliases map 1:1 to their canonical IDs.

Alias	Canonical model
cursor48	anthropic/claude-opus-4.8
cursor47	anthropic/claude-opus-4.7
cursor46	anthropic/claude-sonnet-4.6
cursorgpt5	openai/gpt-5.1
cursorgpt5c	openai/gpt-5-codex
cursorgem	google/gemini-2.5-pro
cursordv4	deepseek/deepseek-v4-pro

First call

Open the chat panel (Cmd/Ctrl-L), pick anthropic/claude-opus-4.8 (or the alias cursor48) from the model dropdown, and send a prompt. Confirm the request landed in the Nimbus dashboard's live Requests view within a second.

Troubleshooting

"OpenAI API Key" field disabled: you're on the free tier. Custom endpoints are Pro-only.
Verify fails with 401: the key is missing the sk-nim- prefix, or Cursor is still using the last-verified endpoint. Toggle Base URL off and back on to re-verify.
Model not in the dropdown: you added the model name but Cursor caches the list — quit Cursor entirely and relaunch.
Composer/Agent uses a Cursor-hosted model instead: Cursor Composer routes some flows through their own backend even when you have a custom key. Explicitly pick a Nimbus model from the model dropdown at the bottom of the Composer panel.
Model responds with "model not found": you added a bare name like claude-opus-4.8. Use the fully-qualified anthropic/claude-opus-4.8 — full catalog at Models.