Two ideas make AI easy to change later:
  • A provider is a company or server that runs models (OpenAI, Anthropic, Google Gemini, Mistral, Groq, or any server that speaks the OpenAI format), together with your key for it.
  • A model is a name your app asks for, like chat or embed, that points at one of a provider’s models.
Because your app only knows the name, you can move chat from one provider’s model to another in Studio and nothing in your code changes.

​
Connect a provider

Open AI → Providers → Add provider, choose one and paste its key. For your own server, enter its address instead. The key you paste is sealed on arrival. Studio only ever shows that a key is stored. To change it, paste a new one. Switching a provider Off stops every model on it.
An OpenAI-compatible address must be reachable from the internet. Addresses on a private network are refused, so a model definition cannot be used to reach machines it should not.

​
Add models

Open AI → Models → Add models. Studio asks the provider which models your key can use and lists them under Chat and Embeddings. Tick one or more. They get simple names (chat, embed, chat-2…) that you can change. If you want a model the provider does not list, such as a fine-tune or a brand-new release, add it by name under Use a model that is not listed. Open a model to change:

​
Who can use a model

Keep models on Servers only unless a browser really needs to call one. A model open to the browser can be used by anyone who can use your app, and the cost is yours.

​
Try a model

AI → Playground lets you chat with any model in Studio, change its settings and see the tokens used, before you write any code.

​
See what it costs

AI → Usage shows calls, tokens, errors, speed and cost for each model over time. Every call is recorded with its model, token counts, time and cost, but never with what was asked or answered.

​
Next

Chat and streaming

Ask a model from your app and stream the answer.

Embeddings

Turn text into vectors.