Providers and Models

Deutsch: Anbieter und Modelle

In AI Studio, a provider is the combination of a service that runs AI models and one of its models, for example, OpenAI with a certain GPT model, or ollama on your computer with a model you downloaded. You can set up as many providers as you like and choose the one that suits each task.

Three kinds of providers

You set up providers in "Settings":

Kind	Used for	Section in Settings
LLM providers	the chat and the assistants	"Configure LLM Providers"
Transcription providers	voice input and transcribing audio and video files	"Configure Transcription Providers"
Embedding providers (beta)	searching your data sources	"Configure Embedding Providers", only when the preview feature for data sources is on

AI Studio needs at least one LLM provider. The other two kinds are only needed for the features they serve.

Adding an LLM provider

Click "Add Provider" and fill in the dialog from top to bottom:

1. "Provider": the service. "Create account" opens its sign-up page.
2. "API Key": the key from your account with the service. For "Self-hosted", the field is called "(Optional) API Key" and appears once you have chosen a host.
3. "Hostname": only for "Self-hosted" and "LiteLLM". The address of the server, starting with http:// or https://, for example, http://localhost:1234.
4. "Host": only for "Self-hosted". The program that runs the model: LM Studio, llama.cpp, ollama, vLLM, or llmman.
5. "Hugging Face Inference Provider": only for "Hugging Face". Choose a provider that runs the model, or let Hugging Face choose the fastest, the cheapest, or one in your preferred order.
6. "Model selection": click "Load models" and choose a model. AI Studio needs the API key for this, or, for a self-hosted model, a host that can list its models. For a few services, you type the name of the model instead, and "Show available models" opens their list of models.
7. "Instance Name": a name of your choice, at most 40 characters and different from your other providers.

Click "Add" to save the provider. To change it later, use "Edit"; the button then reads "Update".

Expert settings

"Show Expert Settings" reveals settings for special cases. Wrong values can break the model, so change them only when you know what the model can do:

• "Override Model Capabilities": tells AI Studio which inputs the model accepts, such as several images, audio, or video, and whether it can call tools. "Reset" goes back to what AI Studio knows about the model.
• "Reasoning (thinking) behavior": whether the model thinks before it answers, from "Automatic" to "Always on".
• "Override Model Limits": the "Context window in tokens", that is, how much text the model can handle at once, and how many "Images" one message may carry. This matters most for models you run yourself, because their limits depend on how they were set up.
• "Additional API parameters": extra settings for the service, such as "temperature": 0.5, in JSON without the surrounding curly brackets.

When the preview feature for data sources is on, you can also choose a tokenizer file for this provider, so that AI Studio counts tokens more accurately.

Managing your providers

The table under "Configured LLM Providers" groups your providers by service. Each row has these actions:

• "Open Dashboard": opens the website of the service, where you see your usage and costs. Not every service has one.
• "Edit": changes the provider.
• "Delete": removes the provider and its stored API key.

Importing a provider

A colleague can share a provider with you as text. Click "Import Provider", paste the text into "Exported configuration", and confirm. AI Studio only reads the values; it never runs the pasted text. The usual dialog then opens, so you can review everything before you save.

An imported provider usually comes without an API key, so enter yours before you save. Your organization may have switched importing off.

Providers of your organization

When your organization manages AI Studio, some providers come from your IT department:

• A lock means "This provider is managed by your organization." You can use the provider, but you can't change or delete it.
• A key icon means "This provider is managed by your organization. You can set your own API key." Click it to enter your personal key. Everything else stays as your organization set it.

Your organization may also prevent you from adding or importing providers. The buttons then show a lock.

Which provider is preselected

AI Studio can select a provider for you, so you don't have to choose it every time:

• For the whole app: "Settings" → "App Options" → "Preselected provider".
• For the chat: in the chat, open "Chat Options" with the gear icon, switch on "Preselect chat options?", and choose the "Preselected provider". There, you also choose whether new chats start with the provider you used last or always with the preselected one.
• For an assistant: in the assistant, open its options with the gear icon, switch on the preselection, for example, "Preselect translator options?", and choose the "Preselected provider".

The choice of an assistant or the chat wins over the one for the whole app. When you have set up only one provider, AI Studio always uses it.

Where your API keys are stored

AI Studio keeps API keys in the credential store of your operating system: the Keychain on macOS, the Credential Manager on Windows, and a password manager with the Secret Service on Linux. Backups of AI Studio leave them out.

When AI Studio can't store or read a key, the dialog shows the reason under the API key field. On Linux, this usually means that your password manager is locked or missing; see Secure storage for API keys.

Transcription providers

For voice input, add a transcription provider under "Configure Transcription Providers" with "Add transcription provider", then choose it in "Settings" → "App Options" → "Select a transcription provider". Without this choice, dictation and transcription stay off.

You can transcribe with OpenAI, Mistral, Fireworks.ai, Groq, GWDG, Helmholtz, LiteLLM, Requesty, Hugging Face, or on your own computer with ollama, vLLM, whisper.cpp, or llmman.

Embedding providers (beta)

Data sources need an embedding provider, which turns your texts into numbers that AI Studio can search. Add one under "Configure Embedding Providers" with "Add Embedding". "Test" lets you try it with a text of your choice.

Embeddings are available from OpenAI, Mistral, Google, Alibaba Cloud, IONOS, GWDG, Helmholtz, OpenRouter, LiteLLM, Requesty, Hugging Face, or on your own computer with LM Studio, ollama, vLLM, or llmman.

Where to go next

• Provider Confidence: how far you trust each provider, and how to keep sensitive work away from some of them.