JetBrains puts Air, a multi-agent workspace, inside its IDEs

JetBrains Air, now in early access, runs several coding agents at once inside JetBrains IDEs, gives each session its own Git worktree and lets you review the results with the IDE’s code analysis tools. Air doesn’t sell models; it’s a workspace for the agents you already use.

In early October, JetBrains released JetBrains Air as a plugin for its IDEs and opened an EAP, an early public preview before the official release. The plugin ships with 2026.3 EAP builds, and on 2026.2 you install it from JetBrains Marketplace. It works in JetBrains IDEs such as IntelliJ IDEA, PyCharm and WebStorm, but not in Android Studio, and it doesn’t need a separate license.

Why a chat panel stopped being enough

Until now, AI in the IDE mostly meant a chat panel on the side of the screen. Once you hand bigger jobs to AI agents, programs that carry out multi-step tasks on their own, and run several at once, a single chat panel makes it hard to track which agent has done what.

Air handles this by managing work per session. It shows progress, diffs, cost and the worktree for each session in one place, so when several agents are running you can see right away which one is spending the most.

Worktree isolation and the IDE’s analysis tools

Each agent session runs in its own Git worktree, a Git feature that checks out branches into separate working folders. When an agent edits code, its changes don’t mix with the folder you’re working in or with other agents’ work. You can open an agent’s changes as a diff straight from the session, review them and leave comments on specific lines.

Agents also use the project information the IDE already has. JetBrains says the IDE’s code analysis resolves usages, dependencies and types across the codebase, so agents spend fewer tokens and make fewer mistakes. A token is a small piece of text, roughly part of a word, that AI models read and write. While agents work, you keep using code insight, refactorings, the debugger and version control as usual.

How it differs from Cursor

Cursor is a separate editor built on VS Code. Air is a plugin for the JetBrains IDE you already use, and it doesn’t sell models. No agents come preinstalled, and AI features stay off until you sign in to an agent.

Instead, Air detects the agents installed on your machine and connects them. According to JetBrains’ product page, it supports Codex, Gemini, GitHub Copilot, Claude Agent and Junie out of the box, and other agents connect through ACP, an open protocol that links coding agents to editors. Secondary reports say Cursor’s agent can be connected too if it’s installed.

A JetBrains AI subscription isn’t required. In most cases you can use the third-party subscription or agent login you already have, and your code then goes straight to that provider without passing through JetBrains.

What’s unconfirmed or easy to misread

• A Claude subscription can’t be used to authorize Claude Agent in Air’s chat view. JetBrains’ help docs suggest using Claude Code in the terminal view instead.
• Cloud runs, which keep going after you close your laptop, currently require a JetBrains AI subscription.
• Junie Lite, the free way to try Air, is offered for a limited time and needs a JetBrains account.
• JetBrains hasn’t said whether Air will be priced separately after the EAP, or at what price.

What to watch

The first thing to watch is how Air’s pricing and cloud-run terms settle once the EAP ends. Limits like the one on using a Claude subscription in the chat view show that how freely third-party agents can plug in will decide how Air competes with AI-first editors like Cursor.

Sources

• A New Agentic Experience: JetBrains Air in IDEs – EAP Now Open (JetBrains Blog)
• JetBrains Air in IDEs overview (JetBrains Help)
• JetBrains Air in IDEs (JetBrains)