What JetBrains Air is and who it fits
JetBrains Air describes itself as one system for building software with agents. Instead of shipping yet another proprietary coding agent, Air is a layer where you pick an agent and a model per task and then review the result. The official lineup is Claude Agent (Anthropic), OpenAI Codex, Gemini CLI (Google) and JetBrains' own Junie, and the Air app can also run any ACP-compatible agent installed on your machine. Junie is now part of JetBrains Air.
Air fits developers and teams who already use more than one coding agent and want a single place to start tasks, track parallel work and review diffs, plus organizations that need central control over which agents and models developers may use.
Surfaces: IDE plugin, browser, CLI and Air Teams
Air is reachable from several surfaces. In JetBrains IDEs you install the plugin and run agents inside the IDE. In the browser at air.jetbrains.cloud you can work from any machine and share sessions with your team. Air Gateway brings terminal agents like Claude Code and Codex into Air from the command line. The product page also shows a mobile surface. Air finds compatible agents on your machine and connects them with a click, and you can add others from the agent registry.
Review is part of the core loop: you open diffs directly from an agent session, validate changes with IDE tools and leave line comments for the agent to act on. Air Context adds semantic indexing so agents start with an understanding of the project rather than scanning the whole codebase.
Cloud tasks and automations
Air Teams runs tasks in containerized cloud environments. You can start them from the browser, from AI Assistant in IntelliJ-based IDEs, or from the Air app by choosing Cloud as the run environment. The repository is cloned shallow, so only the latest commit is present until someone fetches more history. Environments are suspended after 75 minutes of inactivity and archived after 24 hours, while results stay on the task branch. Snapshots can speed up later runs, but JetBrains says to treat preserved state as a speed-up, not a guarantee.
Automations are cloud tasks triggered by schedules or events. Examples on the product page include code review on PR creation, documentation updates on merge and issue triage.
Governance, privacy and data path
For organizations, access is managed in JetBrains Central Console under AI governance. Administrators decide whether developers can use AI at all, set monthly credit limits, choose which agents are available, control whether cloud tasks are allowed, and decide whether connectors reach cloud environments. By default every organization developer has AI access, can use all agents and can run cloud tasks, so admins who want tighter control need to change those defaults. Repository access goes through a JetBrains Air connection app on GitHub, or through instance registration for GitHub Enterprise Server and GitLab Self-Managed.
JetBrains documents cloud tasks as giving stronger isolation, because they do not use your local files, processes or credentials unless you explicitly provide the configuration. As of the documentation dated 30 September 2026, Gemini CLI is temporarily unavailable for cloud tasks.
Pricing and accounts
Air gives you two ways to pay. Local tasks can run on JetBrains AI or on your own account: a Claude subscription (Pro, Max or Team), an Anthropic API key, a ChatGPT Plus, Pro or Team account or OpenAI Platform key for Codex, or a Google account or AI Studio key for Gemini CLI. Cloud tasks and automation runs always use the JetBrains AI provider and spend JetBrains AI credits, even if you connected your own key.
JetBrains AI licensing is credit-based. The AI Free, AI Pro, AI Ultimate and AI Enterprise tiers differ mainly in quota, and each AI Credit equals $1 USD charged in local currency. Top-up credits are valid for 12 months. The product page states that JetBrains AI credits are billed at public API rates. Check the JetBrains AI pricing page for current tier prices before buying.
The Bottom Line
JetBrains Air is a credible answer to agent sprawl. It treats Claude Agent, Codex, Gemini CLI and Junie as interchangeable workers behind one review and governance layer, and it is well documented. Its main trade-offs are that cloud work always bills through JetBrains AI credits, and that its value depends on how many agents you actually run. If you only use one agent, that agent's native app may be simpler. Compare it with the standalone agents it orchestrates, such as Codex, Claude Code and Junie, on aicoolies.
