E2B

E2B provides isolated Linux sandboxes for AI agents and applications that need to execute code, manipulate files or run external tools. Developers create and manage environments through Python or JavaScript SDKs, while a template defines the software and configuration available when each sandbox starts. This separates generated code from the application host and gives the agent a controlled machine with its own filesystem, processes and network surface. E2B also publishes a CLI and API references for lifecycle, command, file and connection operations; the canonical repository is licensed under Apache 2.0.

The platform covers more than one-shot code execution. Sandboxes can expose streamed or background commands, interactive terminals, SSH connections and public URLs, and they can upload or download files, mount volumes, export metrics and carry application metadata. Lifecycle features include persistence, snapshots, forking and auto-resume. Separate guides cover desktop environments for computer-use agents, remote browsers, coding agents, CI/CD jobs and data analysis. Availability, concurrency and compute charges vary by service plan; the current billing documentation is authoritative.

Top features

• Isolated Linux environments: Start a dedicated sandbox for agent-generated code and tools instead of executing them on the application server.
• Python and JavaScript SDKs: Create sandboxes, run commands, inspect results and manage lifecycle from common agent-development stacks.
• Reusable templates: Define base images, packages, users, working directories, environment variables and startup or readiness commands for repeatable environments.
• Command and terminal access: Run foreground or background processes, stream output, open an interactive PTY, reconnect to a running sandbox or use documented SSH access.
• Lifecycle and observability: Work with sandbox metadata, metrics, lifecycle events, webhooks, snapshots, forks, persistence and OpenTelemetry export.

Use cases

• Give coding agents a disposable workspace for editing, dependency installation, tests and build commands without exposing the host machine.
• Run code-interpreter sessions for data analysis, charts and generated notebooks or files.
• Provide a virtual Linux desktop or remote browser to computer-use agents.
• Isolate AI-assisted tests, validation and code review inside CI/CD workflows.
• Prebuild task-specific templates so every agent run starts with the same tools and operating environment.