E2B

What is E2B?

E2B is cloud infrastructure that gives AI agents isolated computers for executing generated code and using real-world tools. Developers create a sandbox through JavaScript or Python SDKs, then let an agent run commands, use a terminal or browser, install packages, inspect errors, and create or upload files without executing untrusted work on the application server.

Sandboxes run in Firecracker microVMs and can be customized with templates, CPU, memory, packages, and system libraries. The platform is model-agnostic and works with different model providers and agent frameworks. Hosted options can start quickly, while enterprise deployment includes customer cloud, VPC, on-premises, or self-hosted arrangements.

How to Use E2B

1. Create an E2B account and obtain a server-side API key.
2. Install the JavaScript or Python sandbox or code-interpreter SDK.
3. Create a sandbox with the required template, resources, and session limit.
4. Give the agent narrowly scoped tools for code, terminal, browser, and file operations.
5. Validate inputs, set timeouts and budgets, and treat generated code as untrusted.
6. Capture logs and results, then terminate the sandbox and retain only intended outputs.

Core Features

• Isolated agent computers: Runs untrusted workflows inside Firecracker microVMs.
• Code interpreter: Executes Python, JavaScript, Ruby, C++, and other Linux-compatible code.
• Terminal and filesystem: Supports commands, packages, uploads, downloads, and generated files.
• Browser tooling: Gives agents a browser inside the sandbox for appropriate workflows.
• Fast startup: Starts same-region hosted sandboxes without traditional long cold starts.
• Custom templates: Prepares reusable environments with selected dependencies and settings.
• Model and framework independence: Works with multiple LLMs and agent frameworks.
• Enterprise deployment: Supports hosted, customer-cloud, VPC, on-premises, and self-hosted patterns.

Use Cases

• AI code execution: Run generated analysis or transformation code away from production hosts.
• Data analysis agents: Process uploaded datasets and return charts or files.
• Computer-use agents: Provide an isolated environment with browser and terminal tools.
• Background coding: Let agents build and test changes in disposable machines.
• Secure MCP tools: Expose sandboxed execution as a controlled agent capability.

Pricing

Hobby has no monthly platform fee and includes a one-time $100 usage credit, one-hour sessions, and up to 20 concurrent sandboxes. Pro costs $150 per month plus usage and adds configurable CPU and RAM, sessions up to 24 hours, and up to 100 concurrent sandboxes. Compute is billed per second; for example, two vCPUs cost $0.000028 per second, with memory charged separately.

Frequently Asked Questions

Does E2B provide the language model?

No. It is model-agnostic sandbox infrastructure used alongside a chosen model.

Can packages be installed at runtime?

Yes. Agents can install packages and system libraries, or developers can create a custom template.

Does isolation remove the need for limits?

No. Applications still need permission boundaries, timeouts, resource budgets, network controls, and output validation.