Skip to main content
With an AI Gateway API key you can use the models your organization gives you in Cursor’s chat and agent. Requests go through AI Gateway like any other: the key only reaches models you’re allowed, and usage counts toward your limit. Your organization needs Allow 3rd party tools & harnesses turned on, and Cursor setup available.

​
Set it up

Create a key in My Library › Models and pick Cursor under Start using this key: it shows everything to paste, with your key filled in. Already have a key? Open Set up Cursor in the same section.
  1. Create an API key. Select Create key, name it (Cursor on my laptop) and copy it. It’s shown once.
  2. Point Cursor at OpenWork. In Cursor, open Settings › Models › API Keys. Turn on OpenAI API Key and paste your key. Turn on Override OpenAI Base URL and paste the base URL from the page. It ends in /api/v1/cursor.
  3. Add your models. Choose a model on the page and copy its name, for example openwork/claude-opus-5.5-k3f9. In Cursor’s Models settings, select Add model and paste it. Repeat for each model you want.
  4. Pick a model and run locally. Open Cursor’s model picker, select one you added, and run the agent on This Mac (Local), not in the Cloud.
Use Cursor’s Local mode. Cursor’s cloud agents don’t work with OpenWork models yet: a prompt sent to the Cloud fails. Choose This Mac under the prompt box before you send.
The base URL must end in /api/v1/cursor, not /api/v1. The Cursor address answers in the formats Cursor expects for every model, including Claude and Gemini models.

​
Model names

Cursor has one field per model, used both as its id and as the name it shows. OpenWork gives each of your models a short name for it:
The four characters tell apart models with the same name, for example the same model from two providers. Copy names from the page rather than typing them.

​
Thinking level

Cursor doesn’t send a thinking level for models added this way, so the level is part of the name. On the page, choose a level next to a model and copy the name it shows:
Without a level, the model uses its provider’s default. To switch between levels in Cursor, add the same model once per level.

​
Good to know

  • Tab completions and Cursor’s own models keep using your Cursor plan.
  • Cursor calls the base URL from its own servers. A self-hosted AI Gateway has to be reachable from the internet for Cursor to use it. If it isn’t, Cursor says the model you selected isn’t available.
  • While the override is on, Cursor sends its OpenAI models to the base URL too. Only the names from the page work there.

​
When it doesn’t work