JetBrains has published a preview of cloud agents in JetBrains Air, enabling AI agents to run remotely in a JetBrains-managed cloud environment with isolation from the user’s local machine.
The preview demonstrates how cloud agents operate in practice: a task started in IntelliJ IDEA can be followed from a browser or a phone, with the agent’s changes reviewed after completion.
Cloud agents are currently available to organizational users via Air in IDEs and Air Teams in the browser, with a broader release planned for individual users later.
HOW CLOUD-BASED SANDBOXING WORKS IN AIR
Agent sandboxing means running the agent inside an isolated and tightly controlled environment that restricts which resources it may access and which actions it may perform. TechStaged has also covered Faster Developers Don’t Make a Faster Team: JetBrains study details AI-assisted productivity gaps.
In Air, each task runs in a cloud VM; the agent itself executes inside a Docker container within that VM. The VM is treated as untrusted and is not reused between tasks.
A proxy inside the VM routes all connections from the agent container, with default restrictions that limit access to a preset list of trusted domains (for example, common source control and package registries such as npm).
Secrets are stored encrypted and exposed to the task as environment variables, with shared secrets usable by teammates without exposing their values, and personal secrets kept private. Tasks respect the developer’s repository permissions.
WHAT IT MEANS FOR SECURITY AND DEPLOYMENT
JetBrains frames cloud execution as providing stronger isolation by default, reducing local setup requirements, and giving organizations more control over network access, environment configuration, and task execution policies.
Two broad execution options exist: local setups that wrap the agent locally and remote cloud-based setups that isolate the agent by default. The remote path uses a cloud VM and a container to limit what the agent can access.
AVAILABILITY AND WHAT HAPPENS NEXT
Organizational users currently have access to cloud agents via Air in IDEs and Air Teams in the browser, with a broader release for individual users planned later.
This approach is part of a broader effort to address safety concerns as AI agents can read files, execute code, use tools, and interact with external services.
RELATED COVERAGE
- Faster Developers Don’t Make a Faster Team: JetBrains study details AI-assisted productivity gaps
- JetBrains Discontinues Swift Language IDE Support in Kotlin Multiplatform Plugin
- JetBrains joins OpenSSF to bolster secure software development
- GitLab’s agent orchestration cuts the wait between development stages with Duo Agent Platform
- Developer Tools articles
SOURCES
- The JetBrains Blog: How Cloud-Based Agent Sandboxing Works in Air Published · Primary source
