llms.txt: machine-readable documentation index. Markdown version of this page is also available.

JetBrains Air Help

Quickstart guide

Follow this guide to install Air in IDEs, start your first agent session, provide context, and review the changes suggested by the agent. For a more detailed walkthrough, watch the video at the end of this guide.

1. Install Air in IDEs

Air in IDEs is a separate plugin that provides the workspace for running coding agents in the IDE. How you set it up depends on the version of your IDE:

The Air in IDEs plugin is not bundled with the IDE. Install the plugin from JetBrains Marketplace:

  1. Open settings or press Ctrl+Alt+S in your IDE and then select Plugins.

  2. Switch to the Marketplace tab.

  3. Search for Air and click Install.

  4. Restart the IDE if prompted.

The Air in IDEs plugin is bundled with the IDE, so you do not need to install it. To make sure that the plugin is enabled:

  1. Open settings or press Ctrl+Alt+S in your IDE and then select Plugins.

  2. In the Bundled section, locate the Air plugin in the list and make sure it is enabled.

2. Open a project

Agent sessions work with the files and configuration of the project that is open in the IDE.

  • Open the project where you want to work with an agent.

3. Select an agent and start a session

Air in IDEs starts a session from a preset – an agent configuration that preselects the model, effort, access level, and other session options. A preset is created automatically for every supported agent that Air in IDEs detects on your machine or that you install.

  • In the main toolbar, either click New Session to immediately start a session with the selected agent, or click and select a specific agent preset from the list. This opens a new session with the agent in a new editor tab.

    Agent widget menu in the main toolbar listing the presets and agent defaults available for a new session, and the Add Agents item.

4. Authorize the agent

To work on a task, an agent needs an available authorization source. You can tell whether the agent picked up your existing configuration or you need to configure one after you send the first message in the session:

  1. Type a message in the prompt field and press Enter.

    If Air in IDEs detects an authorization that the agent already uses on your machine, the agent answers the message, and you can proceed to the next step.

  2. If the agent shows the Choose a sign-in method to continue list instead, click the preferred authorization method or select it with the Up and Down arrow keys, and press Enter.

    Agent session where the first message triggers the sign-in method list with the agent account, JetBrains AI organization workspaces, and Add an AI provider.

  3. Complete the authorization:

    Depending on the agent, you may be prompted to complete the authorization in a browser or terminal.

    1. Provide your login details in the browser or terminal that opens, or follow another authorization flow that the agent provides.

    2. Return to the session.

    Agents that support JetBrains AI authorization offer a JetBrains AI entry if you have an eligible license or organization workspace associated with your JetBrains Account.

    1. Select the JetBrains AI entry from the list.

    2. Wait until the connection to the selected agent is established.

    Use this option to work through a provider subscription of your own. Selecting the Add an AI provider option opens the AI | Air | Accounts settings page:

    1. Click More Providers and select the provider that you want to add.

      Accounts settings with a configured OpenAI API key entry and the More Providers popup listing the available API key types.

    2. In the dialog that opens, enter the credentials that the provider requires, such as an API key, and configure any additional settings.

    3. Click OK to save the settings.

    4. Return to the session and select the configured provider in the list.

After the authorization completes, the agent is ready to work on the task.

5. Configure the agent

If needed, you can configure the agent before you send your first request for the task. The available options depend on the selected agent, but you usually can:

  • Select a processing model and the reasoning effort for the task.

  • Select an operation mode that determines which actions the agent can perform without asking for confirmation.

  • Configure any other settings available for the selected agent.

For more information about presets and their settings, refer to Configure and manage presets.

6. Describe the task and add context

Give the agent a specific goal and enough information to complete it. Describe the expected result, mention important constraints, and attach the project items that the agent needs to inspect.

  1. Type your request in the prompt field using natural language.

  2. Click the button and attach the files, commits, or skills that the agent needs for the task.

    Add-context menu open in the prompt field, listing options to attach project files, screen context, commits, and skills to the request.

For more ways to provide information to the agent, refer to Attach context.

7. Send the prompt and follow the task

When the prompt contains the task and required context, send it to start the agent. The agent then reports its progress and may ask you questions while it works.

  1. Click Send in the bottom-right corner of the prompt field, or press Enter.

  2. Follow the agent's progress and provide any information it requests.

To learn how to continue a conversation, queue prompts, or answer agent questions, refer to Interact with the agent.

8. Review and manage the changes

Review the result before committing the changes. In a code review, you go through the changed files and leave notes on the lines that need rework or an explanation. The agent receives the notes with your next message, and then answers your questions or updates the code.

  1. When the agent finishes the task, click Review Code in the header of the session tab.

    Session tab with the Review Code button and the +94 -0 line counts in the header, above an agent reply and a list of one file changed since the last turn.

  2. Select the changes to review. For the result of your first task, click Local changes to review all uncommitted changes.

    Review Code popup in the session header with Local changes selected under Uncommitted, above a list of the commits of the master branch.

    The code review starts. The Review in progress tab of the Agent Sessions tool window lists the changed files, and the diff of the selected file opens in the Code Review editor tab.

    Review in progress tab with the tree of four changed files, three of them viewed, next to the Code Review tab with the diff of taskFormats.js and the Go to Session button on its toolbar.

  3. Go through the changed files: select a file in the Review in progress tab, or use the buttons on the toolbar of the Code Review tab.

  4. If a change is unclear or needs rework, leave a note: click in the gutter next to the line, type your question or request, and press Enter.

    Code Review tab with a saved note card on line 28 and a note draft on line 32 that shows the Add sorting by due date session with the In code review label.

  5. When you have gone through all files, click Finish & Attach Comments in the Review in progress tab.

    Review in progress tab with the Local + 2 commits scope, nine changed files, two notes on taskManager.js, and the Finish & Attach Comments and Cancel buttons at the bottom.

    The review ends, and the notes appear in the text field of the session as one attachment per file.

    Text field of the session with two attachments: reporting.test.js with one note and taskManager.js with two notes.

  6. Add a prompt if needed, for example Address the notes, and send the message. The agent answers your notes or updates the code.

  7. Repeat the review until the changes suit you, and then commit them.

For more information about reviewing the changes, refer to Review changes with code notes.

Watch the walkthrough

Watch an eight-minute demo that goes beyond this guide. In the video, you learn how to:

  • Choose the right agent for the job.

  • Create separate sessions for different tasks.

  • Track the agent's progress without watching every step.

  • Start a request by pressing Ctrl twice.

  • Switch between chat and terminal sessions.

  • Dictate a prompt with voice input.

  • Tell an unread update from a session that needs your input.

Next steps

06 November 2026
Air in IDEsKey concepts