You can use [Agent Studio](https://console.cloud.google.com/agent-platform/studio/) to design,
test, and manage prompts for Google's
[Gemini](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/google-models.md.txt) large language models (LLMs) and
partner and open models. Agent Studio supports certain partner and open models that are
offered on Agent Platform as
[models as a service (MaaS)](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/use-partner-models.md.txt),
such as Anthropic Claude models and Meta's Llama models.

> [!NOTE]
> **Note:** On your initial use of partner and open models, Agent Platform prompts you to accept the provider's terms and conditions. You must do this once for each partner and open model provider to start using their models.

In this quickstart, you do the following:

- Send these prompts to the Agent Platform API using samples from the generative AI prompt gallery, including the following:
  - A summarization text prompt
  - A code generation prompt
- View the code used to generate the responses

## Sample prompts in Agent Studio

A prompt is a natural language request submitted to a language model that
generates a response. Prompts can contain questions, instructions, contextual
information,
[few-shot examples](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/prompts/introduction-prompt-design.md.txt#few_shot_examples),
and partial input for the model to complete. After the model receives a prompt,
depending on the type of model used, it can generate text, embeddings, code,
images, videos, music, and more.

The sample prompts in [Agent Studio](https://console.cloud.google.com/agent-platform/studio/)
[Prompt Gallery](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/prompt-gallery.md.txt) are predesigned to help
demonstrate model capabilities. Each prompt is preconfigured with specified
model and parameter values so you can open the sample prompt and click
**Submit** to generate a response.

## Test the Gemini model using a summarization text prompt

Send a summarization text prompt to the Agent Platform API. A summarization
task extracts the most important information from text. You can provide
information in the prompt to help the model create a summary, or ask the model
to create a summary on its own.

1. Go to the **Prompt Gallery** page from the Agent Platform
   section in the Google Cloud console.  

   [Go to Prompt Gallery](https://console.cloud.google.com/agent-platform/studio/prompt-gallery)

2. In the **Tasks** drop-down menu, select **Summarize**.

3. Open the **Audio Summarization** card.

   This sample prompt includes an audio file and requests a summary of the file
   contents in a bulleted list.
4. Notice that in the settings panel, the model's default value is set to
   **gemini-3.7-flash** . You can choose a different Gemini model
   in the **Model settings** panel.

5. Click **Submit** to generate the
   summary.

   The output is displayed in the response.
6. To view the Gemini Enterprise Agent Platform code used to generate the transcript
   summary, click **Code**.

   In the **Get code** panel, you can choose your preferred language to get the
   sample code for the prompt, or you can open the Python code in a
   Colab Enterprise notebook.

## Test the Gemini model using a code generation prompt

Send a code generation prompt to the Agent Platform API. A code generation task
generates code using a natural language description.

1. Go to the **Prompt Gallery** page from the Agent Platform
   section in the Google Cloud console.  

   [Go to Prompt Gallery](https://console.cloud.google.com/agent-platform/studio/prompt-gallery)

2. In the **Tasks** drop-down menu, select **Code**.

3. Open the **Generate code from comments** card.

   This sample prompt includes a [system instruction](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/prompts/system-instruction-introduction.md.txt)
   that tells the model how to respond and some incomplete Java methods.
4. Notice that in the settings panel, the model's default value is set to
   **gemini-3.7-flash** . You can choose a different Gemini model
   in the **Model settings** panel.

5. To complete each method by generating code in the areas marked
   `<WRITE CODE HERE>`, click **Submit** .

   The output is displayed in the response.
6. To view the Gemini Enterprise Agent Platform code used to generate the code,
   click **Code**.

   In the **Get code** panel, you can choose your preferred language to get the
   sample code for the prompt, or you can open the Python code in a
   Colab Enterprise notebook.

## Automatic intent detection in the prompt workspace

When chatting or testing prompts in Agent Studio, background intent detection
automatically evaluates each submitted message to identify whether you intend
to build an autonomous agent or an application. If an intent is detected, a
contextual banner appears with a one-click action to transition your request to
the appropriate builder.

### How intent detection works

Each time you submit a message in the chat prompt workspace, a background
evaluation processes your input:

- Agent intent detected: A banner appears with a **Create agent** action.

  - If the detection engine has enough context to determine what kind of agent
    you want to build, it directs you to Agent Designer and automatically submits
    your seed prompt to initiate building the agent immediately.

  - The canvas opens with generated instructions, system configurations, and
    pre-attached tools ready for further editing.

- App intent detected: A banner appears with a **Create app** action.

  - Clicking the action button navigates to the App Builder interface.

  - The prompt field is pre-populated with your prompt details if context is
    available.

  - Unlike the agent flow, App Builder never automatically submits the build;
    it stops on the page so you can review and manually submit the prompt.

### Billing and pricing

Each background intent detection evaluation is charged to your Google Cloud
account:

- Model used: Intent detection is powered by Gemini 3.5 Flash-Lite, an
  optimized lightweight model.

- Cost: Because the model is lightweight, the charge per detection is
  minimal. Approximately $0.002 per message evaluation.

- Standard inference charges apply separately to your primary prompt and
  response.

### Create an agent using intent detection

To create an agent from a chat session using automatic intent detection:

1. In the Google Cloud console,  

   [Go to Agent Platform](https://console.cloud.google.com/agent-platform/agent-platform/studio/)

2. In the navigation menu, click **Studio** \> **+ New** \> **Chat** \>
   **New chat**.

3. In the prompt field at the bottom of the workspace, enter a prompt that
   clearly states your intent to build an agent, along with a description of the
   agent's purpose and tasks.

   > [!NOTE]
   > **Note:** The banner only appears when you express a clear intent to build an agent.

4. Click **Submit** .

5. When the response begins, locate the intent detection banner above the
   prompt field.

6. In the banner, click **Create agent**.

7. In the confirmation dialog, click to continue. Agent Studio opens the visual
   canvas in **Agent Designer** and initializes your agent flow.

8. In the side panel, review and configure the populated settings:

   - **Details**: Edit the agent name and description.

   - **Instructions**: Review and adjust the system instructions.

   - **Tools**: Verify and configure attached tools (such as Google Search).

### Create an app using intent detection

To create an application from a chat session using automatic intent detection:

1. In the Google Cloud console,  

   [Go to Agent Platform](https://console.cloud.google.com/agent-platform/agent-platform/studio/)

2. In the navigation menu, click **Studio** \> **+ New** \> **Chat** \>
   **New chat**.

3. In the prompt field at the bottom of the workspace, enter a prompt that
   clearly states your intent to build an application, along with a description
   of the application.

   > [!NOTE]
   > **Note:** The banner only appears when you express a clear intent to build an app.

4. Click **Submit** .

5. When the intent detection banner appears above the prompt field,
   click **Create app** . Agent Studio opens the **App Builder** workspace with
   your prompt pre-populated in the input field.

6. Review and edit the prompt text as needed.

7. Inside the prompt field, click
   **Submit** to start the application
   build process.

### Disable intent detection

If you want to prevent background evaluations and avoid per-message detection
charges while chatting:

1. In the Google Cloud console,  

   [Go to Agent Platform](https://console.cloud.google.com/agent-platform/agent-platform/studio/)

2. In the navigation menu, click **Studio** \> **+ New** \> **Chat** \>
   **New chat**.

3. In the workspace header, click **Settings** (gear icon).

4. In the settings menu, locate the **Detect build intent** setting.

5. Click the toggle switch to the off position.

When disabled, no background detection runs, suggestion banners don't appear,
and no detection fees are charged.

## What's next

Overview

### [Build overview](https://docs.cloud.google.com/gemini-enterprise-agent-platform/build.md.txt)


Learn how to build agents in Google Agent Platform.

Guide

### [Agent Studio capabilities](https://docs.cloud.google.com/gemini-enterprise-agent-platform/agent-studio/agent-studio-capabilities.md.txt)


Learn about the capabilities of Agent Studio.

Guide

### [Deploy your prompt as a web application](https://docs.cloud.google.com/gemini-enterprise-agent-platform/agent-studio/deploy-vais-prompt.md.txt)


Follow the guide to deploy your prompt as a web application from Agent Studio.