Quick start
JetBrains Air is an Agentic Development Environment where you delegate coding tasks to AI agents and stay in control of the workflow. In this quickstart, you install JetBrains Air, sign in to an agent provider, open a project, run your first task, and review the result.
1. Install Air
Download JetBrains Air for macOS from the official page at air.dev. Support for Windows and Linux is planned for 2026.
2. Sign in to an agent provider
To run tasks, JetBrains Air needs at least one connected agent provider. On the first launch, you are prompted to sign in.
If you have a JetBrains AI subscription, connect JetBrains AI. It works as a universal provider and gives you access to Claude Agent, OpenAI Codex, Gemini CLI, and Junie under a single subscription.
If you prefer to use a specific provider account – Anthropic, OpenAI, or Google – sign in to that account directly. You can also bring your own API key (BYOK).
To connect additional providers later, open . For details, see Set up Air.
3. Open a project
After you sign in, open the project you want to work on. In JetBrains Air, each project opens as a workspace that groups the agent sessions, Git state, and tools for that project in one place.
Open a project
To open a local project, click Open and select the project folder on your computer.
To clone a project from Git, click Clone from Git, enter the repository URL in the Source URL field, and choose a location on your computer.
When the workspace opens, JetBrains Air asks whether you trust the code in the folder. Click Trust only if you trust the authors of the project – opening the project may run scripts or import code, which can execute anything in the project. If you want to look around first, click Preview: JetBrains Air functionality will be limited but safer.
4. Run your first task
In JetBrains Air, you work with the agent through the Chat tool. You describe the goal, add the context the agent needs, and send the task.
Run your first task
Open the Chat tool and describe the goal in the task input.
State what you want to achieve and what done means – expected behavior, constraints, and known failures or error messages. The clearer the goal, the better the result.
Add context for the task.
Attach the files, symbols, or code the agent needs so it does not have to guess:
Type
@or click
(Add context) to attach files, folders, Git items, terminal output, or mention symbols such as classes, functions, or methods.
Select code in the editor and click Add to Task to attach the selection.
For the full list of ways to add context, see Task context.
Adjust how the agent runs. In the task toolbar, you can:
Pick a different agent and model, a permission mode that controls how freely the agent acts.
If you want the agent to draft an implementation plan before it changes code, select the Plan permission mode. See Plan mode.
Choose a run environment:
Local Workspace – the agent runs on your machine and writes directly to your working copy.
Git Worktree – the agent works in a separate Git worktree on a dedicated branch, so your main working copy stays untouched until you apply the result.
Docker – the agent runs in an isolated container with its own filesystem, useful for running untrusted code or keeping the environment clean.
Send the task.
Press ⏎ or click Send. The agent works on the task while you keep editing or start another task in parallel.
Learn more about running and managing tasks in Run tasks.
5. Review the changes
When the task is done, JetBrains Air shows the Changes button in the Chat tool. Click it to open the diff in the Task Changes tab and decide what to do next.
Review with an agent
For a fast first pass, use Review with Agent. JetBrains Air runs a separate review task with a fresh agent session that checks the diff against your review focus areas and leaves comments on specific lines. Accept the comments you agree with and send them back to the main task – the main agent uses them as follow-up instructions and updates the code.
For the full workflow, including how to review another scope or change the review agent and prompt, see Review with agent.
Review the diff yourself
You can also review the diff manually. In Task Changes, you can:
switch between Unified and Side-by-side views and navigate between hunks
comment on specific lines and send the comments back to the task as review feedback
select code in the diff and click Add to Task to attach it to a follow-up instruction
For details, see Review changes.
6. Apply the changes
When you are satisfied with the result, apply the changes to your local workspace and integrate them into your normal workflow. How you do this depends on the run environment you picked for the task.
From the local workspace
If you ran the task in Local Workspace, the agent already wrote the changes to your working copy. Open Task Changes, enter a commit message, and click Commit. Then push the commit to the remote repository.
From an isolated environment
If you ran the task in Git Worktree or Docker, the agent worked outside your local working copy. Bring the result back to your local workspace in one of two ways:
Click Apply Locally to copy the changes into your current working copy as uncommitted changes. Then commit and push as usual.
From the Apply Locally drop-down, select Checkout Branch Locally to check out the task branch in your working copy. Push the branch or open a pull request as usual.
For the full workflow – including how to revert changes or roll back to a previous commit – see Accept changes.
Where to go next
JetBrains Air offers more features that you can layer on as you grow into the agentic workflow.
Plan mode – let the agent draft an implementation plan before it changes code, and approve it before execution.
Permission modes – control when the agent can edit files or run commands without asking.
Project instructions – commit project-level rules (
CLAUDE.md,AGENTS.md) that shape how agents work in your repository.MCP servers – connect external tools and data sources to the agent through the Model Context Protocol.
Skills and Commands – reuse and compose instructions and shortcuts for recurring task patterns.
Multitasking – run several tasks in parallel and switch between them.