# Google AI Studio model connection

Google AI Studio connects supported Gemini models to your AI features.

*Release 3.3 · admin, developer*

**Settings › AI › Models** · **Settings › AI › Accounts**

## What it is

**Google AI Studio** lets an AI feature use a supported Gemini model. INSIDE sends the feature's conversation and reads the model's reply. The connection supports both complete replies and streaming.

## Who uses it and why

Administrators agree where Gemini should be used. Developers check required inputs, functions and reply formats with Powerfront. Powerfront configures the model and account for you.

## Where to find it

**Google AI Studio** is a **Provider** choice in **Settings › AI › Models**. Powerfront configures its **Model**, **API Key** and **Endpoint Base** in **Settings › AI › Accounts**. Both screens are Powerfront-managed.

## How it works

1. An **AiFlow** account selects a model configured for **Google AI Studio**.
2. INSIDE sends the feature's instructions and conversation. A direct connection can also send available functions, function-selection controls and thinking options.
3. The model returns text or requests a function. INSIDE retains the association between tool calls and results for later turns.
4. Where Google requires signed conversation history, INSIDE preserves it with the relevant turns. This private continuation information is excluded from the public reply representation.

Requests can include image and file content, and can ask for JSON or a reply that follows a specified structure. The selected model must support the requested input and format.

## Configure it

Ask Powerfront to create the Google model record and account and supply the **API Key**. Select that account for the required AI feature in **Settings › AI › Features**, using the **Developer** role or the **Settings** role with **AI** access. An unset **Endpoint Base** uses the built-in Google v1beta model service.

On a direct connection, **Reasoning Effort** requests these thinking budgets from compatible models:

| Choice | Requested budget in tokens |
|---|---:|
| None | 0 |
| Low | 1,024 |
| Medium | 8,192 |
| High | 16,384 |
| XHigh | 24,576 |

**Default** sends no explicit thinking budget. Ask Powerfront to check the model accepts the chosen option.

### Settings

| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Reasoning Effort | Settings › AI › Accounts | website | Not set: the model service decides | Powerfront, on request | On a direct Google connection, requests a thinking budget from compatible models; Default leaves the provider's behaviour unchanged. |
| Endpoint Base | Settings › AI › Accounts | website | Not set | Powerfront, on request | Overrides the Google model-service address; an unset value uses the built-in Google v1beta model service. |

### Fields

| Field | Where | Required | What it is for |
|---|---|---|---|
| Provider | Settings › AI › Models | Yes | Select Google AI Studio when preparing a Gemini model record. |
| Model | Settings › AI › Accounts | Yes | Selects the configured Google model for an AiFlow account. |
| API Key | Settings › AI › Accounts | Yes | Supplies the Google service credential; a non-empty key is required for Google requests. |

## Limits and good to know

- Google requests require a non-empty **API Key**, including requests routed through the gateway in this release.
- Gateway requests have reduced support in 3.3. They omit new function declarations, function-selection controls and thinking configuration, and send instructions as conversation text. Ask Powerfront to use a direct connection where those capabilities are needed.
- A narrow compatibility retry can change a Google v1 request to v1beta after an error. This is not a switch to another model or provider.
- Streaming availability does not determine how a particular operator or visitor screen displays the answer.

## Depends on and works with

### This feature uses

- **Requires** [AI accounts and shared defaults](AIP-001.md): An AI account supplies Google model selection and credentials.
- **Configured in** [AI reply streaming settings](AI-037.md): The AI feature's streaming options control whether supported incremental output is delivered progressively.
- **Configured in** [AI model catalogue](AIP-002.md): Powerfront prepares a model with Google AI Studio as its provider.
- **Works better with** [AI functions and required inputs](AI-016.md): Direct connections can use functions exposed by the AI feature.
- **Works better with** [Rich content in AI replies](AI-026.md): Enabled rich replies let compatible Gemini models return validated structured content.
- **Works better with** [Images and attachments for AI agents](AI-039.md): Requests can include image and file content when supported by the model.
- **Works better with** [Cloudflare AI Gateway connection](AIP-007.md): Google requests can use a gateway, with reduced tool and thinking support in this release.

### Used by

- **Enhances** [AI model provider selection](AIP-003.md): Google AI Studio supplies another model connection.
- **Enhances** [Cloudflare AI Gateway connection](AIP-007.md): Google AI Studio requests can use the gateway with release-specific capability limits.

## FAQ and troubleshooting

### Can I set up Google AI Studio myself?

Powerfront manages the model and account screens. Ask Powerfront to prepare the connection, then use the account in your AI feature.

### Why do functions work directly but not through the gateway?

The 3.3 gateway path omits new function declarations and function-selection controls. Ask Powerfront to check whether the feature needs a direct Google connection.

### Why does Reasoning Effort have no effect?

The model must accept thinking budgets, and the connection must send them. The Google gateway path in this release does not send that configuration.

### Must I manage Google's signed conversation history?

No. INSIDE retains the private information needed for the relevant conversation turns.