E2B

End-user uploads a target CSV or dataset via the main application frontend.

The application's LLM agent reviews the data and writes a visualization script.

The script is routed to an isolated E2B Data Analysis Sandbox for safe execution.

The generated chart and processed text outputs are securely relayed back to the user's dashboard.

The training orchestrator triggers an evaluation epoch.

E2B provisions tens of thousands of isolated sandboxes concurrently.

Each sandbox executes a unique AI-generated code snippet against predefined unit tests.

Pass/fail metrics are aggregated and returned to the trainer as a reward signal.

User assigns a high-level UI task to an agent (e.g., 'Book a flight using a browser').

E2B provisions a Desktop Sandbox with a complete virtual OS.

The agent interacts with the sandbox visually, moving the mouse and typing.

Upon task completion, the resulting state or file is extracted and the sandbox is destroyed.

Coding agent is assigned a GitHub issue.

Agent starts an E2B Code Sandbox, accesses the internet, and clones the target repo.

Agent modifies the code, utilizes the sandbox terminal to run local tests, and reads the logs.

Once tests pass, the agent commits the changes and pushes them remotely.

User prompts the agent to compile a comprehensive market report.

The agent spawns a Deep Research Sandbox and writes dynamic scraping scripts.

The sandbox executes the scripts, navigating rate limits and headless browsers securely.

Raw data is synthesized within the sandbox and a clean final report is generated.