Browser
A headless Chromium instance. Agents can use three automation approaches, alone or combined within a single workflow:- DOM-based interaction — click, type, and select elements through Playwright selectors.
- Playwright scripts — run deterministic JavaScript against the live page for precise, repeatable steps.
- Computer Use (CUA) — enable the
computer_usecapability on an AI Task node so the model reasons over screenshots and issues coordinate-based actions when DOM tooling is not enough.
Linux
A full Linux desktop session. Use this when the workload goes beyond a web browser — for example interacting with native applications, terminals, or file managers. Linux environments use Computer Use (CUA) models that reason over screenshots and produce mouse/keyboard actions to control the desktop.Windows
A Windows desktop session for workloads that require Windows-only applications such as legacy enterprise software or desktop clients. Like Linux, Windows environments rely on Computer Use (CUA) models for all interaction.Shared conventions
Regardless of environment type, every sandbox uses the same/home/agent directory layout (shared/, workspace/, downloads/, uploads/), so file handling and tool behaviour stay consistent. See Files for details.
Files
Directory layout and file API access
AI Capabilities
Tools and capability identifiers
Playwright node
Deterministic browser scripts
Profiles
Credentials and environment configuration