E2B provides isolated cloud sandboxes for agents to run code. When the agent needs to execute Python, install a library or process a file, execution happens on a disposable machine kept apart from the user’s system.
What it is for
- Running model-generated code without risk to the local machine
- Giving the agent an environment with a filesystem, terminal and controlled internet
- Processing data, generating charts and running scripts in analysis pipelines
- Evaluating agents that must truly act, not just return text
How it works
- The sandbox is created from a Python or JavaScript SDK and starts in milliseconds
- The environment has a persistent filesystem during the session and supports file uploads
- You can install packages, expose ports and follow execution in real time
- Long sessions keep the sandbox alive between agent calls
- The core is open source and can be self-hosted; the cloud handles operations
Availability
- Open-source core under the Apache-2.0 licence
- Commercial cloud service with a limited free plan and billing per sandbox time
Points to note
- Model-generated code needs network, CPU, memory and time limits
- Long-lived sessions generate continuous cost
- For simple tasks, your own container may be enough

