JetBrains Air

Product Introduction

1. Definition: JetBrains Air is a specialized Agentic Development Environment (ADE) designed to orchestrate and manage multiple autonomous AI coding agents within a unified, code-aware workspace. It serves as a dedicated desktop application that abstracts the complexities of AI task execution, providing a structured workflow for agent-driven development.

2. Core Value Proposition: JetBrains Air exists to transform AI-assisted coding from "one-shot" prompt engineering into a reliable, high-velocity "agentic loop." By enabling the parallel execution of multiple agents—such as Codex, Claude Agent, and Gemini CLI—within isolated environments, Air solves the bottlenecks of sequential task processing and manual context management. It leverages JetBrains' proprietary code intelligence to ensure agents have precise project context, leading to higher accuracy in code generation and refactoring.

Main Features

1. Multi-Agent Parallel Orchestration: Air allows developers to launch and manage multiple AI agents (Anthropic’s Claude Agent, OpenAI’s Codex, Google’s Gemini CLI, and JetBrains’ Junie) simultaneously. Unlike standard chat interfaces, Air manages these agents as independent task loops, allowing them to work on different parts of a codebase concurrently without resource contention or interference.

2. Isolated Execution Environments (Docker & Git Worktrees): To ensure stability and security, Air executes agent tasks in isolation. It utilizes Docker containers or Git worktrees to create sandboxed environments where agents can run tests, build projects, and modify code. This architecture prevents "hallucinated" or experimental code from affecting the main development branch until the user explicitly reviews and commits the changes.

3. Code-Aware Context Definition: Air utilizes language-aware navigation and indexing to provide agents with deep project context. Users can define tasks by referencing specific symbols, classes, methods, commits, or images. This metadata-rich approach ensures that agents understand the relationships between different files and modules, significantly reducing errors compared to agents that rely on raw text-based file uploads.

4. Async Task Monitoring and Oversight: The platform provides a centralized dashboard for tracking the progress of asynchronous agent tasks. Developers can switch between different agent workspaces to provide real-time input, course-correct logic, or review intermediate diffs, ensuring that the human developer remains the ultimate decision-maker in the agentic loop.

Problems Solved

1. Pain Point: The Sequential Processing Bottleneck: Traditional AI coding tools typically handle one request at a time, forcing developers to wait for completion before starting the next task. JetBrains Air solves this by allowing parallel execution, enabling a "Manager" workflow where the human delegates multiple sub-tasks to different agents.

2. Target Audience:

• Senior Software Engineers: Seeking to delegate boilerplate, unit testing, or refactoring tasks across large-scale microservices.
• Full-Stack Developers: Working across unfamiliar languages or frameworks who need AI agents to bootstrap CLI tools or APIs.
• DevOps Engineers: Automating infrastructure-as-code or script generation in isolated environments.
• AI-First Engineering Teams: Organizations transitioning from simple AI autocomplete to autonomous agentic workflows.

3. Use Cases:

• Cross-Language Prototyping: Rapidly building a CLI in a language like Go or Kotlin while leveraging agents to handle syntax and library selection.
• Large-Scale Refactoring: Assigning separate agents to update different modules of a project simultaneously via Git worktrees.
• Automated Bug Hunting: Running agents in parallel to explore project history and symbols to identify and patch vulnerabilities in isolation.

Unique Advantages

1. Differentiation (Orchestration vs. Chat): While most AI tools focus on the underlying Large Language Model (LLM), JetBrains Air focuses on the workflow surrounding the model. It replaces the manual process of "juggling terminals and pasting file paths" with an integrated workspace that understands project structure and maintains session history.

2. Key Innovation (Agent Agnosticism & BYOK): Air is uniquely flexible, allowing users to choose between using their JetBrains AI subscription or "Bringing Your Own Key" (BYOK) for Anthropic, OpenAI, or Google APIs. This ensures that developers are not locked into a single model provider and can always use the most effective agent for a specific technical challenge.

Frequently Asked Questions (FAQ)

1. How is JetBrains Air different from a standard AI IDE plugin? JetBrains Air is a standalone Agentic Development Environment (ADE) focused on orchestration, whereas IDE plugins typically focus on autocomplete and inline chat. Air handles the execution lifecycle—including environment isolation via Docker and Git worktrees—allowing agents to run complex, multi-step tasks independently of your primary coding window.

2. What agents are currently supported in JetBrains Air? At launch, JetBrains Air supports Claude Agent (Anthropic), Codex (OpenAI), Gemini CLI (Google), and Junie (JetBrains). The platform is designed to be agent-agnostic, allowing for the addition of new models and specialized coding agents as the ecosystem evolves.

3. Does JetBrains Air require a paid subscription? Air is currently free for macOS (Windows and Linux versions coming soon). Users with a JetBrains AI Pro or Ultimate subscription can use all supported agents at no additional cost. Alternatively, users can connect their own API keys (BYOK) from providers like OpenAI or Anthropic and pay those providers directly.

4. How does Air prevent AI agents from breaking my local code? Air uses "Isolated Execution" through Docker containers and Git worktrees. Every task an agent performs happens in a separate environment, meaning agents can run, test, and modify code without causing branch collisions or polluting your main local development environment. You only merge the changes once you have reviewed them using Air's code-intelligent review tools.