API keys and models
Connecting OpenRouter or a lab directly, choosing which models appear, and running a model on your own computer.
On this page
Which key#
- OpenRouter. One key reaches every provider's models. It is the simplest way to start, and OpenRouter reports what each answer cost.
- Anthropic, OpenAI, Google, xAI, or Meta. A key from the lab itself. Answers go straight to the lab. If you have both a lab key and an OpenRouter key, Use an AI company's own key first decides which is used.
- A local model server. Anything that speaks the OpenAI chat format, such as Ollama. Give its address and its models appear like any other.
Paste the key under Settings → Developer → AI keys and press Test key. One tiny request confirms it works.
Keys are sealed with a server secret before they are stored. A copy of the database on its own cannot read them.
Which models appear#
Settings → Models lists the models you can choose from anywhere in the app, with the price per million tokens read and written, and how each is reached. Star up to three to have them picked for you by default. Add any model by its id, such as meta/muse-spark-1.3.
The lab engines offered for AI visibility are the ones on this list that can search the web. Any model reached through OpenRouter can search, as can OpenAI, Anthropic, Google, and xAI models reached with their own keys. Perplexity's Sonar comes through OpenRouter.
Advanced settings#
Under Settings → AI columns you can set how many rows an AI column tries first, how many run at once, the model used for compressing sources, the fallback model used when the main one is overloaded, and how long to wait for an answer.
Where keys are used#
An assistant uses the model set under Configure. A check, a comparison, and a visibility check use the models you choose for them. The plan's allowances apply only to calls on our keys. See Plans and usage.
Updated Nov 7, 2026