Quick start
JetBrains Air is the Agentic Development Environment where OpenAI Codex, Claude Agent, Gemini CLI, and Junie execute independent task loops without interfering with one another. You define a task, point it to the right parts of your project, and review the result with the tools you need in one place.
JetBrains Air focuses on agent orchestration without replacing your existing development workflow. Use JetBrains Air for agent-driven work, and use your IDE for the rest.
To get started, open a project, create your first task, and review the proposed changes. You can run tasks locally by default, or isolate them in Docker containers and Git worktrees when you want sandboxing and concurrent work.

How to try JetBrains Air
You can download JetBrains Air for macOS from the official page at air.dev. Support for Windows and Linux is planned for 2026.
First run
When you first run JetBrains Air, you will be prompted to log in to your Claude, OpenAI, Google Gemini, and JetBrains account.

Open a project
To open an existing local project, click Open and navigate to the project folder on your computer.
To clone a project from Git:
Click Clone from Git.
In the Source URL field, enter the URL of the Git repository.
In the Location field, specify the path where you want to store the cloned repository.

When the workspace opens, JetBrains Air will ask whether you trust the code in the folder. If you click Preview, JetBrains Air functionality will be limited.
For all features to work, JetBrains Air may need to execute project code. Actions like importing the project, running scripts, and executing Git commands can run malicious code if the source is untrusted. Only trust the project if you trust the code authors.

If you trust the code in the folder, click Trust.
You can now start working.
Define a new task
Defining a task involves writing a clear prompt and providing all necessary context. When you work with complex tasks, specific instructions help produce accurate results. In JetBrains Air, you define the task through an interactive chat mode. This is not a single-shot request. You can refine the task step by step.
You can use the interactive plan mode to ask the agent to create a detailed execution plan. You can also guide the agent by asking questions, adding context, and clarifying requirements during the conversation. This helps the agent understand your intent and produce better results.
JetBrains Air currently works with Claude Agent, OpenAI Codex, Gemini CLI. Make sure you have an active Anthropic, OpenAI ChatGPT, or Google Gemini subscription.
Configure task settings
Before running the task, you can configure the execution environment, AI model, and permissions.
JetBrains Air supports three execution environments:
Select the execution environment:
Git Worktree: creates a separate working branch of the repository. This mode provides isolation from your main branch while still using your local environment. Note that you may need to reinstall project dependencies or repeat location-specific setup for every new task.
Docker: runs in an isolated container. You must have Docker Desktop installed and a valid license to use it. This mode offers complete isolation for code changes and tools. Isolation means that all edits, commands, and dependencies stay inside the container and do not affect your local workspace or system environment.
Local Workspace: runs directly in your current workspace. This mode has the fastest startup and uses your existing environment, but changes are applied to your project folder. It does not provide isolation.

Click the Agent and Model drop-down menu and select the AI model you want to use.

Select a Permission Mode to control how the agent asks for approval before editing files or running commands. Permission mode options depend on agent type.
Ask Permission: prompts for permission on the first use of each tool.
Auto-Edit: automatically accepts file edit permissions.
Plan: analyzes code without editing files or running commands.
Full Access: skips all permission prompts.

Add context to the task
You can add context to the task prompt in several ways:
Click the Add context button to attach specific items. You can add:
Files and Folders: reference files and folders from your workspace.
Git Branches: reference a specific branch.
Git Commits: reference a specific commit.
Local Changes: reference changed files from the History tool.
MCP Server: connect an MCP server to provide extra tools.
Terminals: reference a specific terminal tab.
Upload from Computer: upload files from your computer.

Add symbol mentions through the @ menu. Use the Symbols submenu to reference symbols from the current file or from the project.
Select code in the editor and click Add to Task on the floating toolbar to attach the selected code.

Add a code snippet manually by typing it inside triple backticks (
```).
Run and manage async agents
JetBrains Air can run multiple tasks asynchronously. For example, you can start one task adding new tests and an agent fixes a bug in another task. You can continue to work on a new feature at the same time.

View the list of tasks
To view the list of tasks in your current project, press Command+1. You can also click the Tasks menu icon.

Add a new task
Press Command+\ and click the New Task button.
Click the New Task icon (
).
A task might have the following states:
Running: the agent is working on the task. You can follow the task progress and provide additional input at any moment.

Input required: the agent has paused and needs information or a decision from you.

Done: the agent has completed the task. You can review and commit the changes.

Canceled: you stopped the task before the agent completed it.

Tasks completed before today are moved to a dedicated section.

Review changes
When a task is finished, switch to the Changes tab to review the changes in the files. The diff panel shows the changes made by the agent.
You can provide feedback to the agent directly in the diff panel, similar to a regular code review. Use the icon in the gutter to comment on a specific line or selection.

Provide feedback to an agent
Navigate to .
Select the type of diff to display. You can choose between:
Select the code that you want to change and click Add comment in the popup.
Type your prompt for the agent. The comment is added to the chat as part of a future task.

Push your changes
After your review, JetBrains Air helps you bring your changes back to the local workspace, and you can then push the changes to the remote repository.
Commit and push your changes
Navigate to .
In the Changes tool, select the files that you want to commit.
In the commit message field, type a description of the changes.
Click Commit.
Navigate to . Alternatively, click the History icon on the tools panel.
In the History tool, click Push. If this is your first push, JetBrains Air may ask for a GitHub token. Click the Generate link to create a token on GitHub. The minimum required permissions are set automatically, and you do not need to change them.
Other features
JetBrains Air includes other features that help you work with agents:
Built-in web preview for verifying results in web applications.
MCP server integration
JetBrains Air can interact with external tools and data sources by using the Model Context Protocol (MCP). By connecting to MCP servers, JetBrains Air gains access to a range of tools that significantly extend its capabilities.
Configure an MCP server
Press Command+, to open settings and select .
Click Add Global MCP Server and paste your server configuration in JSON format. Example:
{ "mcpServers": { "alphavantage": { "command": "uvx", "args": ["av-mcp", "YOUR_API_KEY"] } } }The server appears in the list.
Manage MCP server connections
Press Command+, to open settings and select .
When your server appears in the list, click the … button to Edit, Restart, Disable, or Delete the server connection.

View available MCP server tools
Click the tool link next to your server name to see the list of available tools.

Adding Claude Agent
JetBrains Air supports Claude Agent built-in commands and custom commands that you can trigger by typing / in a task prompt.

You can connect Claude Agent by using Log in via Anthropic Console (API Billing).
Connect an Anthropic account
Press Command+, to open settings and select .
Click Connect next to Anthropic.
In the login dialog, click Log in via Anthropic Console (API Billing).
Complete authentication in the browser window that opens. After successful login, Anthropic appears in the list of connected providers.

Select a Claude Agent model
After you connect Anthropic, you can select a Claude Agent model for tasks.
Choose a Claude Agent model
Open a new task. For more information, refer to Define a new task.
Click the Agent and Model selector in the task toolbar.
In the Claude Agent Model list, select a model.

Configure Claude Agent
Press Command+, to open settings and select .
You can configure the following Claude Agent settings:
To configure global settings, click Edit next to Global Settings. Enter settings for your Claude Agent instance in JSON format. For more information, see Claude Agent settings.

Adding OpenAI Codex agent
JetBrains Air supports OpenAI Codex models for running agentic tasks. You can connect your OpenAI account and select a Codex model directly from the JetBrains Air interface.
JetBrains Air supports two types of OpenAI accounts:
ChatGPT account (Plus, Pro, or Team), which uses your existing ChatGPT subscription.
OpenAI Platform account, which uses API billing.
Connect an OpenAI account
Log in with OpenAI
Press Command+, to open settings and select .
Click Connect next to OpenAI.
In the login dialog, select the account type you already use:
ChatGPT Account (Plus, Pro, Team)
OpenAI Platform (API Billing)
Complete authentication in the browser window that opens. After successful login, OpenAI appears in the list of connected providers.

Select an OpenAI Codex model
After connecting your OpenAI account, you can select an OpenAI Codex model, which JetBrains Air will use for tasks.
Choose a Codex model
Open a new task. For more information, refer to Define a new task.
Click the Agent and Model selector in the task toolbar.
From the list, select one of the available OpenAI Codex models, for example:
GPT 5.3 Codex
GPT 5.2 Codex
GPT 5.1 Codex Max
GPT 5.1 Codex Mini
GPT 5.2

Adding Gemini CLI agent
JetBrains Air supports Gemini CLI models for running agentic tasks. You can connect a Gemini CLI provider and select a model directly in JetBrains Air.
JetBrains Air supports the following types of Gemini CLI accounts:
Google Account (Personal or Workspace)
Google AI Studio (API Billing)
If you use Google Cloud Platform, select Log in with Vertex AI.
Connect a Gemini CLI provider
Connect Gemini CLI
Press Command+, to open settings and select .
In the list of providers, find Gemini CLI and click Connect.
In the login dialog, select the account type you already use:
Google Account (Personal or Workspace)
Google AI Studio (API Billing)
If you use Google Cloud Platform, click Log in with Vertex AI.
Complete authentication in the browser window that opens. After successful login, Gemini CLI appears in the list of connected providers.

Select a Gemini CLI model
After you connect Gemini CLI, select the model for your task.
Choose a Gemini CLI model
Open a new task. For more information, refer to Define a new task.
Click the Agent and Model selector in the task toolbar.
In the Gemini CLI Model list, select a model, for example:
Gemini 3.1 Pro
Gemini 3 Pro
Gemini 3 Flash
Gemini 2.5 Pro
Gemini 2.5 Flash
Gemini 2.5 Flash Lite

Adding Junie agent
JetBrains Air supports Junie models for running agentic tasks. You can connect a Junie provider and select a model directly in JetBrains Air.
You can add Junie by logging in to your JetBrains account.
Connect Junie
Press Command+, to open settings and select .
In the list of providers, find Junie and click Connect.
Complete authentication in the browser window that opens. After successful login, Junie appears in the list of connected providers.

Select a Junie model
After you connect Junie, select the model for your task.
Choose a Junie model
Open a new task. For more information, refer to Define a new task.
Click the Agent and Model selector in the task toolbar.
In the Junie Model list, select a model, for example:
Gemini 3 Pro
Gemini 3 Flash
GPT 5.2 Codex
GPT 5.2
GPT 5
Grok 4.1 (Fast)
Opus 4.5
Opus 4.5 (Fast)
Sonnet 4.5
Web preview
The preview feature in JetBrains Air lets you instantly preview your web application.

The preview opens automatically for web-based projects when your server starts, or for static projects that contain an HTML file. By default, the preview pane opens in preview mode, indicated by the toolbar in the upper-right corner.
If the preview does not appear, try running it manually.
Run the preview manually
Press Command+Shift+K and start typing New Preview Tool. Select New Preview Tool.
You can switch from the preview to the source code view to inspect the underlying HTML source. The code is read-only, indicated by a lock icon. You cannot edit the code in this mode.
Switch between preview and source code modes
Click the web or code mode icon in the upper-right corner of the Preview tool.

If the page does not open or shows an error, you can ask the chat to summarize the problem and suggest possible fixes. You can also use the Reload Page button in the toolbar to refresh the preview and load the latest changes.
By default, the preview is in responsive mode, shown by the Expand icon in the Devices toolbar. Responsive mode uses the full width and height of the editor and adjusts automatically as you resize the window.

