Key takeaways
- Air replaces Fleet as JetBrains' answer to agentic coding — built on Fleet's codebase, launched December 2025
- Three isolation modes (Docker, Git worktrees, local) with cloud execution coming — most flexible isolation of any leader
- Requires your own Anthropic or OpenAI subscription — JetBrains hosts nothing, you bring your own key
FAQ
What is JetBrains Air?
Air is an agentic development environment (ADE) that lets developers delegate coding tasks to AI agents running in parallel, with IDE-like code review for merging results.
How much does Air cost?
Air itself is free during preview. You need an Anthropic (Claude) or OpenAI (ChatGPT) subscription to power the agents.
What agents does Air support?
Currently Claude Agent and OpenAI Codex. JetBrains' Junie and Google Gemini are on the roadmap.
What platforms does Air support?
macOS only during preview. Windows, Linux, and browser-based versions planned for 2026.
Who competes with Air?
Cursor, Claude for Mac, Codex App, Warp Oz, and Emdash are the main competitors in the Mac coding agent app category.
Executive Summary
Air is JetBrains' agentic development environment (ADE) that lets developers delegate coding tasks to AI agents working in parallel while maintaining full control through IDE-like code review. Built on the discontinued Fleet codebase, Air launched in December 2025 as JetBrains' response to the wave of AI-native coding tools like Cursor and Warp Oz.
| Attribute | Value |
|---|---|
| Company | JetBrains |
| Founded | 2000 (Air launched 2025) |
| Funding | Bootstrapped ($800M+ annual revenue) |
| Employees | ~2,000 |
| Headquarters | Prague, Czech Republic |
Product Overview
Air represents a fundamental shift in how JetBrains approaches AI-assisted development. Rather than adding AI features to existing IDEs like IntelliJ IDEA or PyCharm, Air is purpose-built for a workflow where the developer guides and reviews rather than writes code directly.
The core premise: delegate complex tasks to AI agents that work asynchronously, potentially on multiple tasks simultaneously, then review their output in a dedicated diff interface before committing changes. This positions Air for developers comfortable with significant AI autonomy — closer to code review than pair programming.
Key Capabilities
| Capability | Description |
|---|---|
| Parallel Agent Execution | Run multiple agents on different tasks simultaneously |
| Multi-Mode Isolation | Docker containers, Git worktrees, or local workspace per task |
| IDE-Like Code Review | Diff panel with inline comments and feedback to agents |
| Interactive Task Definition | Refine prompts through conversation, not single-shot requests |
| Local History Snapshots | Roll back entire workspace to pre-agent state |
| MCP Server Integration | Connect external tools via Model Context Protocol |
Product Surfaces / Editions
| Surface | Description | Availability |
|---|---|---|
| macOS Desktop App | Primary client, downloadable from air.dev | Preview (Dec 2025) |
| Windows Desktop | Native Windows client | Planned 2026 |
| Linux Desktop | Native Linux client | Planned 2026 |
| Browser Client | Air running in the cloud | Planned 2026 |
Technical Architecture
Air inherits Fleet's architecture — a modern codebase built from scratch, separate from the 25-year-old IntelliJ platform. This gives it performance characteristics closer to VS Code than traditional JetBrains IDEs, though some users report memory concerns similar to Fleet.
Isolation Modes
Air offers three execution environments for agent tasks:
| Mode | Isolation Level | Startup Speed | Best For |
|---|---|---|---|
| Git Worktree | Medium — separate branch, shared local env | Fast | Most development tasks |
| Docker | High — containerized, nothing affects host | Medium | Untrusted code, experiments |
| Local Workspace | None — changes apply directly | Instant | Quick fixes, trusted agents |
Cloud execution is on the roadmap — agents would continue working even when your machine is off.
Agent Support
| Agent | Status | Requires |
|---|---|---|
| Claude Agent | ✅ Supported | Anthropic subscription |
| OpenAI Codex | ✅ Supported | OpenAI ChatGPT subscription |
| Junie (JetBrains) | 🔜 Coming | TBD |
| Gemini | 🔜 Coming | TBD |
Air uses a BYOK (Bring Your Own Key) model — you authenticate directly with Anthropic or OpenAI. JetBrains doesn't proxy your API calls or charge for AI usage.
Permission Modes
| Mode | Description |
|---|---|
| Ask Permission | Prompts before file edits or commands |
| Auto-Edit | Automatically accepts file changes |
| Plan | Analyzes without editing (read-only) |
| Full Access | No prompts, maximum autonomy |
Strengths
-
JetBrains Backing — Unlike indie tools, Air comes from a profitable company with $800M+ annual revenue and 25 years of developer tooling experience. No VC runway concerns.
-
Most Flexible Isolation — The only leader offering Docker, Git worktrees, AND local workspace options per task. Pick the right isolation level for each job.
-
IDE-Like Review Experience — Code review interface with inline comments feels familiar to developers. Feedback loops back into agent prompts for refinement.
-
BYOK Model — No additional AI costs beyond your existing Anthropic/OpenAI subscription. No per-seat AI pricing.
-
Agent Client Protocol (ACP) — JetBrains is co-developing the open ACP standard with Zed, which could enable broader agent compatibility across IDEs.
Cautions
-
Preview Stability Risks — JetBrains discontinued Fleet after 4 years in preview without ever reaching GA. Air inherits some of that uncertainty.
-
macOS Only (For Now) — Windows and Linux users must wait until 2026. Browser-based access also deferred.
-
No Issue Tracker Integration — Unlike Emdash (Linear, Jira, GitHub) or Warp Oz (Slack, Linear), Air lacks workflow integrations. Tasks are manually defined.
-
Requires External Subscription — Unlike bundled options like Cursor's included models, Air needs a separate Anthropic or OpenAI account.
-
Fleet's Memory Footprint — Some users report Air inherits Fleet's RAM consumption issues, with one noting "doing nothing takes up 1GB of RAM."
Pricing & Licensing
| Tier | Price | Includes |
|---|---|---|
| Preview | Free | Full Air functionality |
| Post-GA | TBD | Pricing not announced |
External costs:
- Anthropic Claude subscription (~$20/month for Pro)
- OR OpenAI ChatGPT subscription (~$20/month for Plus)
Hidden costs: None currently. JetBrains is absorbing Air development costs during preview.
Competitive Positioning
Direct Competitors
| Competitor | Air's Differentiation |
|---|---|
| Cursor | Air offers Docker isolation and IDE-like review; Cursor has broader adoption (50%+ Fortune 500) |
| Claude for Mac | Air is agent-agnostic (supports Codex too); Claude for Mac is Anthropic-only |
| Codex App | Air supports Claude + Codex; Codex App is OpenAI-locked but has polished cloud execution |
| Warp Oz | Air is local-first; Warp Oz has Slack/Linear integration and cloud-first approach |
| Emdash | Air has Docker isolation; Emdash has 20+ agent support and issue tracker integration |
When to Choose Air Over Alternatives
- Choose Air when: You want JetBrains backing, need Docker isolation, or prefer IDE-like code review
- Choose Cursor when: Enterprise adoption and VS Code familiarity matter most
- Choose Claude for Mac when: You're all-in on Anthropic and want first-party integration
- Choose Warp Oz when: Slack/Linear integration is essential
- Choose Emdash when: You need maximum agent flexibility (20+) or issue tracker integration
Ideal Customer Profile
Best fit:
- JetBrains/IntelliJ users wanting familiar paradigms in an agentic tool
- Developers comfortable with significant AI autonomy
- Teams needing Docker isolation for untrusted agent execution
- Users already paying for Anthropic or OpenAI subscriptions
Poor fit:
- Developers needing Windows/Linux support today
- Teams requiring issue tracker integration (Linear, Jira)
- Users wanting everything in one subscription (no BYOK)
- Organizations needing enterprise SSO/SAML (not yet available)
Viability Assessment
| Factor | Assessment |
|---|---|
| Financial Health | Strong — $800M+ revenue, bootstrapped, profitable |
| Market Position | Challenger — new entrant against established tools |
| Innovation Pace | Moderate — building on Fleet foundation, not greenfield |
| Community/Ecosystem | Growing — ACP partnership with Zed promising |
| Long-term Outlook | Cautiously Positive — JetBrains has resources but Fleet's failure raises questions |
JetBrains' financial stability is unquestionable — they're a profitable, bootstrapped company with 25 years of history. The concern is execution: Fleet, Space, Aqua, and Writerside have all been discontinued in recent years. Air represents JetBrains' bet that agentic development is the future, but the company's track record with new product categories is mixed.
Bottom Line
Air is a credible entry from a stable company, offering the most flexible isolation options of any market leader and a familiar review-first workflow. JetBrains' financial health removes the VC runway concerns that affect many competitors.
Recommended for: JetBrains loyalists, developers needing Docker isolation, users wanting BYOK flexibility with Claude or Codex
Not recommended for: Teams needing cross-platform support today, developers requiring issue tracker integration, organizations wanting bundled AI pricing
Outlook: Air has resources and a differentiated approach (isolation flexibility, IDE-like review). Success depends on whether JetBrains can sustain focus and avoid the fate of Fleet. The ACP partnership with Zed suggests strategic thinking about ecosystem rather than going it alone. Watch for GA announcement and enterprise features — if those arrive, Air could challenge Cursor for enterprise adoption.
Research by Ry Walker Research • methodology
Disclosure: Author is CEO of Tembo, which offers agent orchestration as an alternative to individual developer tools.