Code with Gemini Code Assist

As an early-stage technology, Gemini can generate output that seems plausible but is factually incorrect. We recommend that you validate all output from Gemini before you use it. For more information, see Gemini for Google Cloud and responsible AI.

This document describes how you can use Gemini Code Assist, an AI-powered collaborator in Google Cloud, to help you do the following in Cloud Shell:

• Provide guidance to help you solve problems with your code.

• Generate code for your project.

• Receive inline suggestions while you're coding.

Gemini doesn't use your prompts or its responses as data to train its models. For more information, see How Gemini for Google Cloud uses your data.

To help you comply with any license requirements for your code, Gemini Code Assist provides source citations when its suggestions directly quote at length from a specific source. To learn more about how and when Gemini cites sources, see How Gemini helps you generate code and cites sources.

This document is intended for developers of all skill levels. It assumes you have working knowledge of the Cloud Shell Editor and are familiar with Google Cloud. If you prefer, you can also explore Gemini Code Assist in Cloud Workstations, Cloud Code for IntelliJ, and Cloud Code for VS Code.

Connect to Google Cloud and select a project

In this section, you'll connect to Google Cloud and select a Google Cloud project with the Cloud AI Companion API enabled in the Cloud Shell Editor.

If you select a Google Cloud project without the Cloud AI Companion API enabled, you receive an error notification and are instructed to contact your administrator. For more information, see Set up Gemini Code Assist for a project.

To connect to Google Cloud in the Cloud Shell Editor, follow these steps:

1. Launch the Cloud Shell Editor.

2. If you're using the Legacy Editor, click Try the new Editor.

3. In the status bar, click Cloud Code - Sign In.

   

4. If prompted to authorize Cloud Shell Editor to make Google Cloud API calls, click Authorize.

5. Follow the prompts to sign into your Google Account.

   You're now connected to Google Cloud.

Next, to select a Google Cloud project that has the Cloud AI Companion API enabled, follow these steps:

1. In the Gemini Code Assist status bar, click Gemini Code Assist.

   

2. In the Gemini Code Assist menu, select Select Gemini Code Assist project.

3. Select a Google Cloud project that has the Cloud AI Companion API enabled.

   Gemini is ready to use.

   

To test the capabilities of Gemini Code Assist, open an application, or create a sample application in the next section.

Optional: Create a sample application

If you prefer to use an existing application to test the capabilities of Gemini Code Assist, you can skip this section. Otherwise, to create a sample application, follow these steps:

1. In the Cloud Shell Editor, open the command palette: either press Control+Shift+P (for Windows and Linux) or Command+Shift+P (for macOS), and then run Cloud Code: New Application.

2. Select Kubernetes application.

3. Select the Python (Flask): Guestbook application template.

4. Save the new application in your preferred location.

   A notification confirms that your application was created, and a new window opens with your application loaded.

Chat with Gemini Code Assist

In this section, you learn how to open the Gemini Code Assist pane and chat with Gemini Code Assist to get an explanation of your existing code.

To begin chatting with Gemini, follow these steps:

1. Create a new or use an existing code file. If you're using the Python (Flask) sample, you can perform this task in your front.py file: navigate to Explorer > src > frontend and open the front.py file.

2. In the activity bar of the Cloud Shell Editor, click chat_spark Gemini Code Assist.

3. In the Gemini Code Assist pane, enter the prompt Explain this code to me and click send Send.

   Gemini uses the code in your code file as a reference to your prompt and responds with an explanation of your code.

   To refer to a specific block of code instead of all the code in the file, you can select the block in your code file and then prompt Gemini.

Reset chat history

Gemini Code Assist uses the chat history for additional context when responding to your prompts.

If your chat history is no longer relevant to what you're trying to achieve, you can reset the chat history: in the Gemini Code Assist pane, click delete Reset Chat.

Note: The chat history automatically resets when you switch the workspace or exit the Cloud Shell Editor.

Generate code with prompts

The following sections show you how to use Gemini Code Assist to generate code with the example prompt # Function to create a Cloud Storage bucket inside a Python file. You can also select a part of your code and then prompt Gemini Code Assist for help through the chat feature, and receive and accept or reject code suggestions while you code.

Prompt Gemini Code Assist in a code file

1. Create a new or use an existing code file. If you're using the Python (Flask) sample, you can do this in your front.py file: navigate to Explorer > src > frontend and open the front.py file.

2. In your code file, on a new line, enter # Function to create a Cloud Storage bucket, and then press Enter (for Windows and Linux) or Return (for macOS).

3. To generate code, press Control+Enter (for Windows and Linux) or Control+Return (for macOS).

   Next to your prompt text in the Python file, Gemini Code Assist generates the code in the form of ghost text.

4. Optional: To accept the generated code, press Tab.

Optional: Change keyboard shortcut for generating code

If the default keyboard shortcut for generating code isn't working as outlined in the previous section, you can change the keyboard shortcut by following these steps:

1. In the Cloud Shell Editor, click menu Menu, and then navigate to File > Preferences > Keyboard Shortcuts.

2. In the list of keyboard shortcuts, search for Cloud Code: Generate Code.

3. Click Cloud Code: Generate Code, and then click edit Change Keybinding.

4. In the dialog that appears, enter your own shortcut.

5. Press Enter (for Windows and Linux) or Return (for macOS).

   You can now use your newly assigned keyboard shortcut to generate code with Gemini.

Prompt Gemini Code Assist with selected code using chat

Gemini Code Assist can perform tasks or answer your questions based on the code that you select. To get generated code that's based on a prompt with selected code, follow these steps:

1. In the Cloud Shell Editor, open a file in your project that contains code, or use the same code file that you used in previous steps.

2. In the activity bar, click chat_spark Gemini Code Assist to open the Gemini Code Assist pane.

3. In your code file, select a block of code.

4. In the Gemini Code Assist pane text field, enter a prompt for the selected code.

   For example, select a function in your code and enter the prompt Write a unit test for this function:

   

   Gemini uses your selected code as reference and responds to your prompt.

Get inline suggestions while you code

While you write code, Gemini Code Assist makes inline code suggestions that you can either accept or ignore. To try this feature, follow these steps:

1. Create a new or use an existing code file. If you're using the Python (Flask) sample, you can do this in your front.py file: navigate to Explorer > src > frontend and open the front.py file.

2. In your code file, on a new line, start writing a function. For example, if you're in a Python file, write def.

   Gemini suggests code in the form of ghost text.

3. To accept the code suggestion from Gemini Code Assist, press Tab. Otherwise, to ignore the suggestion, press Esc or continue writing your code.

Optional: Disable inline suggestions

If you prefer to disable inline suggestions in Gemini Code Assist, follow these steps:

1. In the Cloud Shell Editor, select menu Menu, and then navigate to File > Preferences > Settings.

2. On the User tab of the Settings dialog, navigate to Extensions > Cloud Code.

3. Scroll until you find the Cloudcode: Gemini Code Assist: Inline Suggestions: Enable Auto list, and then select Off.

   This turns off the inline suggestions. You can still press Control+Enter (for Windows and Linux) or Control+Return (for macOS) to manually trigger inline suggestions.

Disable code suggestions that match cited sources

Gemini provides citation information when it directly quotes at length from another source, such as existing open source code. For more information, see How and when Gemini cites sources.

To prevent code that matches cited sources from being suggested, you can modify the cloudcode.duetAI.recitation.maxCitedLength setting to 0 in a settings.json file:

"cloudcode.duetAI.recitation.maxCitedLength": 0

Use smart actions

To help you be more productive while minimizing context switching, Gemini Code Assist provides AI-powered smart actions directly in your code editor. When you select your code in your code editor, you can view and select from a list of actions relevant to your context.

To use smart actions in your code, follow these steps:

1. In your code file, select a block of code.

2. Next to the selected code block, click lightbulb Show Code Actions.

   

3. Select an action such as Generate unit tests.

   Gemini generates a response that's based on the action you selected.

Preview: Use Code Transformations

Preview

This product or feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of the Service Specific Terms. Pre-GA products and features are available "as is" and might have limited support. For more information, see the launch stage descriptions.

Gemini Code Assist provides Code Transformations, which allows you to select a portion of your code file and request modifications to your existing code through prompts in natural language, using an inline text box.

With Code Transformations, you can do the following:

• Generate comment lines to document your code
• Troubleshoot code with issues
• Improve code readability
• Make code more efficient

You can also view context sources of a generated response in the Gemini: Chat pane.

Prompt Gemini Code Assist with the inline text box

The inline text box allows you to interact with Gemini Code Assist directly in your code file.

To open the inline text box and prompt Gemini, follow these steps:

1. In your code file, press Control+I (for Windows and Linux) or Command+I (for macOS).

2. Enter a prompt, like Write a function to create a Cloud Storage bucket, and press Enter (for Windows and Linux) or Return (for macOS).

   Gemini Code Assist generates the code based on your prompt.

View diff of refactored code

With the diff view built into the inline text box, you can see the changes that Gemini suggests and your original code. You can then accept or discard the suggested changes.

To view the diff of refactored code from the inline text box, perform the following steps:

1. In your code file, select a function and open the inline text box by pressing Control+I (for Windows and Linux) or Command+I (for macOS).

2. Enter a prompt, like Add comments and press Enter (for Windows and Linux) or Return (for macOS).

   When Gemini generates the code changes to your selected function, the diff view appears.

   

3. To accept or discard the changes, click Accept or Discard.

View context sources of a generated response

The context display feature lists your project files (the context sources) that were used as reference to generate responses to your prompts. Knowing which files were used as references helps you to refine your prompts and get more specific responses. Context sources are shown every time you use Gemini chat.

To view the context sources of the generated response, in the Gemini: Chat pane, scroll until you see the Context Sources snippet. Expand the snippet to see all of the context sources.

Use smart commands

Smart commands are commands that you can use in the inline text box to perform specific actions to your code.

To get an explanation of your entire code file with the /explain smart command, follow these steps:

1. In your code file, open the inline text box by pressing Control+I (for Windows and Linux) or Command+I (for macOS).

2. Press / to view the list of smart commands and their descriptions.

3. Select /explain.

   Gemini generates an explanation of the entire code file in the inline text box.

4. Optional: If you want to view the generated response in the Gemini: Chat pane, click View in Chat.

View prompt history

Accessing your prompt history saves you time when you need to enter the same prompt(s) in a large code file.

To access your prompt history, in your code file, open the inline text box by pressing Control+I (for Windows and Linux) or Command+I (for macOS) and then press Control+Up arrow (for Windows and Linux) or Command+Up arrow (for macOS). This allows you to cycle through your older prompts and re-use them when needed.

Enhance code readability for non-native English speakers

If you're a non-native English speaker that relies on translation applications, you can use the inline text box to generate comments in other languages.

To convert English comments to another language or add new comments in another language, follow these steps:

1. In your code file, select code that you want to add comments to.

2. Open the inline text box by pressing Control+I (for Windows and Linux) or Command+I (for macOS).

3. Enter a prompt, like Add spanish comments and press Enter (for Windows and Linux) or Return (for macOS).

   Gemini generates comments in the language that you specified in your prompt and views the diff of these suggestions.

4. To accept or discard these changes to your code, click Accept or Discard.

Test other example prompts

After reading through the Generate code with prompts section of this document, try some of the following example prompts.

Get an explanation of code

1. In your code file, select a function that you want explained.

2. In the Gemini Code Assist pane, enter the prompt Explain this code to me.

   Gemini uses your selected code as a reference and responds with an explanation of the selected function.

Generate test plans

1. In your code file, select the code that you want to add unit tests for.
2. In the Gemini Code Assist pane, enter the prompt Write unit tests for my code.

Get help with debugging code

1. In your code file, select the code that you want to debug.
2. In the Gemini Code Assist pane, enter the prompt Help me debug my code.

Make your code more readable

1. In your code file, select the code that you want to make more readable.

2. In the Gemini Code Assist pane, enter the prompt Make my code more readable.

   If you prefer to focus on a specific part of your code, then select the preferred part of code before prompting Gemini.

Known issues

The following sections outline the known issues of Gemini Code Assist.

License recitation warnings don't persist across sessions

If license recitation warnings don't persist across sessions, refer to the persistent logs:

1. Click menu Menu > View > Output.

2. Select Gemini Code Assist - Citations.

Connectivity issues in the Gemini Code Assist output window

If you see a connection error or other connectivity problems in the Gemini Code Assist output window, try the following:

• Configure your firewall to allow access to oauth2.googleapis.com and cloudaicompanion.googleapis.com.

• Configure your firewall to allow communication over HTTP/2, which gRPC uses.

You can use the grpc-health-probe tool to test connectivity. A successful check results in the following output:

$ grpc-health-probe -addr cloudaicompanion.googleapis.com:443 -tls error: this server does not implement the grpc health protocol (grpc.health.v1.Health): GRPC target method can't be resolved

An unsuccessful check results in the following output:

timeout: failed to connect service "cloudaicompanion.googleapis.com:443" within 1s

To obtain more details, run the following before grpc-health-probe:

export GRPC_GO_LOG_SEVERITY_LEVEL=info

Leave feedback

To leave feedback of your experience, follow these steps:

1. In the status bar, click Gemini Code Assist, and then in the Quick Pick menu, select Send feedback.

2. In the form, fill out the Title and Comments fields.

3. If you want to share your Skaffold or AI Companion logs, make sure that you select the Send Skaffold logs or Send AI Companion logs option.

4. Click Submit Feedback.

What's next

• Learn how to write better prompts.