Settings: AI Models
The AI Models tab is where you connect Erizos Studio to an AI provider. Once a provider is enabled, the Code Assist panel in UI Builder can answer questions about the UI you are building, and propose changes to its script and to its components.
Where to Find It
Navigate to: Studio → Settings → Global Settings → AI Models.
Note
AI Models sits under Global Settings. When Studio is connected to a Studio Server, Global Settings is only listed for administrators. The settings apply to the whole system, not to a single user.
Configuring a provider
Each provider has its own card, with a switch in the card title. All five are off by default, and a card shows its fields only while it is switched on.
- Model - the model to use. Pick one from the list; the list is filtered as you type.
- API Key - the key issued by the provider. It is masked as you type.
- Endpoint - the URL Studio posts the request to. The label and the default differ per provider, see the table below. If it is not a valid
httporhttpsURL, Code Assist reports AI endpoint must be a valid HTTP or HTTPS URL when it is used.
ChatGPT has two further optional fields:
- Organization - an OpenAI organization ID, sent as the
OpenAI-Organizationheader. - Project - an OpenAI project ID, sent as the
OpenAI-Projectheader.
The settings are saved with the rest of the Global Settings, and take effect immediately - no restart is needed.
The providers
| Provider | Endpoint field | Default endpoint | Default model |
|---|---|---|---|
| ChatGPT | Chat Completions Endpoint | https://api.openai.com/v1/chat/completions |
GPT-5.5 |
| Claude | Messages Endpoint | https://api.anthropic.com/v1/messages |
Claude Opus 4.1 |
| Gemini | API Base Endpoint | https://generativelanguage.googleapis.com/v1beta |
Gemini 3 Pro Preview |
| Grok | Chat Completions Endpoint | https://api.x.ai/v1/chat/completions |
Grok 4.20 Reasoning |
| Ollama | Chat Endpoint | http://localhost:11434/api/chat |
Llama 3.2 |
Ollama runs the model locally, on the Studio server machine, and is the one provider with no API Key field - the field is not shown and no key is required. The default endpoint points at the Ollama service on the same machine.
Note
More than one provider can be enabled at the same time. Code Assist uses the first enabled provider that also has a model set, in this order: ChatGPT, Claude, Gemini, Grok, Ollama.
Warning
Requests made with the API key are billed by the provider to the account the key belongs to. The key is stored encrypted in the Studio settings file, but anyone who can open the Global Settings can replace it.
What the AI is used for
Code Assist is the only feature that uses these settings. It lives in the UI Builder script editor, behind the Code Assist button in the editor toolbar; click it to open the assistant panel beside the code.
Ask it for a change in plain language and it answers in the panel. When its answer includes an edit it does not apply it silently:
- a change to the script is shown as a side-by-side diff in the editor, with Accept and Reject buttons in the panel;
- a change to the UI itself is listed in the panel as Reviewing UI changes (n operations), with the same Accept and Reject buttons. Accepting replaces the UI's component tree with the proposed one.
The panel header shows which provider and model are answering - for example Claude: claude-opus-4-1-20250805 - or No model enabled.
The request is sent from the Studio server, not from your browser, so the API key never reaches the client. Studio sends the assistant the last eight messages of the conversation, the script module you have open and the UI's component tree.
Info
If no provider is enabled with a model set, Code Assist shows No AI provider is enabled. Open Settings > AI Models and enable a provider. The server also rejects the request if the chosen provider has no API key (except Ollama).