E2B - The Missing Puzzles Piece For AI Agents

AI Code Execution: Building the Future of Agentic Systems with e2b

In a recent interview on AI Tinkerers' global stage, Vosk, founder and CEO of e2b, shared insights about his company's groundbreaking work in enabling secure execution of AI-generated code. This technology serves as a critical foundation for the emerging world of agentic systems by providing a secure runtime environment for AI-powered applications.

 What is e2b?

At its core, e2b is an open-source developer tool designed to run AI-generated code securely in the cloud. The entire technology stack is fully open source, making it accessible to developers building various AI applications.

The platform consists of two main components:

- **Infrastructure repository**: Focuses on security through "sandboxes" (micro virtual machines) where AI-generated code runs safely

- **SDKs and CLI tools**: Makes it easy to create sandboxes, control them, and connect them to language models (LLMs) or agents

How e2b Works

When developers use e2b, they create a sandbox with a single line of code using either JavaScript or Python SDKs. Each sandbox is essentially a small computer in the cloud dedicated to a specific user session. This architecture allows for hundreds of thousands of simultaneous user sessions, each with its own isolated environment.

The typical workflow involves:

1. Setting up the e2b library

2. Defining tools for the LLM to use (like "execute_python")

3. Creating a sandbox where code can execute safely

4. Running AI-generated code within that environment

Most impressively, these sandboxes spin up in approximately 150 milliseconds, with the team working to reduce this to under 100 milliseconds. This speed is crucial for applications like Perplexity, one of e2b's high-profile customers, where user experience depends on rapid response times.

 Use Cases

Vosk outlined several key use cases for e2b:

1. **Business data analysis and visualization**: Using AI-generated code to analyze and visualize business data, similar to what OpenAI's Code Interpreter offers but customizable

   

2. **Generative UI**: Creating dynamic user interfaces on-the-fly based on user requests, which has improved significantly with newer models like Claude Sonnet

3. **Internal applications**: Building streamlit apps and other tools for internal company use

4. **Reasoning enhancement**: Using code execution to improve LLM reasoning capabilities

5. **API integration**: Having agents connect various services by generating code that works with their APIs

 The Artifacts Demo

One particularly interesting showcase was artifacts.e2b.dev, an open-source demonstration similar to Anthropic's artifacts feature. This NextJS application allows users to:

- Choose from various LLM models

- Generate complete, functional applications

- View and interact with the generated applications in real-time

In the demo, Vosk showed how the system could generate a Streamlit app displaying fictional Airbnb data on a Seattle map, with the entire application running inside an e2b sandbox.

Security and Future Direction

A significant portion of the discussion focused on security considerations when running AI-generated code. Vosk acknowledged that while e2b provides the secure sandbox environment, developers currently need to implement their own permission systems for database access and other sensitive operations.

The e2b team is actively working on an observability layer that would alert users when AI-generated code attempts to access databases or perform potentially sensitive operations, allowing for permission requests similar to mobile operating systems.

?

Looking ahead, e2b aims to:

- Become the default choice for running AI-generated code

- Expand capabilities to include deploying AI-generated code

- Develop checkpoint systems allowing for sandbox forking and parallel execution paths

- Continue inspiring developers with new use cases and applications

 The Team Behind e2b

Originally from Prague, Czech Republic, the e2b team is now based in San Francisco. The founding team has deep technical roots, with some members having known each other since high school. Their shared background in mathematics and computer science has helped them create a solution that addresses a critical need in the AI ecosystem.

By providing this essential infrastructure layer, e2b is enabling developers to focus on building innovative AI applications without worrying about the complexities and security concerns of executing AI-generated code.