E2B: The AI Agents Cloud

Where did you grow up and how did you find out about computers?

Vasek: Tomas (my co-founder) and I grew up in the Czech Republic. I’m from a small town called Uherske Hradiste.

I think my first computer was my dad’s IBM laptop. My main use of it was figuring out how to start Doom and Prince of Persia on the DOS system. Most of my childhood I was interested in animal medicine and physics, not computers. I spent more summers helping my dad at our veterinary clinic, assisting at surgeries rather than hacking.

How did you and Tomas meet and what was your startup journey together?

Vasek: We’ve known each other for over 10 years now, we first met at the start of high school. We initially started being interested in board games, then min-maxing Eve Online, and then started being really interested in how to make video games.  We started learning Python together but very soon moved to building games on iPhone with Unity and C#, and then general apps with Objective-C. Soon we realized our passion was in developer tools and solving our own problems. This led to summer-long hackathons building devtools. We were first working on a product called Devbook, but eventually pivoted to E2B early in 2023.

How are customers using the product today, and what has the initial response looked like from developers?

Vasek: Tens of thousands of developers have used E2B and we are growing quickly: in April we had created 20,000 sandboxes for our customers. In September, we created 2 million of them, and we’ve kept growing every week since. The use cases are varied, but mostly building things that are on the edge of AI. For example, HuggingFace is using E2B to do reinforcement learning on their Open R1 model and LMArena, the most popular leaderboard for AI models, uses our sandboxes to run their Web-Arena evaluations.

We also have enterprise customers like Perplexity and PGA. For them, the most frequent use cases are around AI data analysis and visualizations. The LLM generates code and you need to run it somewhere in the cloud in an isolated environment - that’s our Sandbox. We then send the results back to you, ready to display to customers. The new Perplexity charts are a good example of it:

What is your vision for “The AI Agents Cloud” and where are you starting?

Infrastructure for code interpreting is just the first step for us. We’re building a cloud runtime for AI agents and that’s our building block for the new cloud - Cloud 2.0. Last year we launched Desktop Sandbox - a sandbox that you can connect to any LLMs to control a cloud computer with a graphical environment; this enables our customers to go beyond APIs and code, and interact with web apps as well. We also get a lot of usage for our Fragments app - our open-source version of Claude Artifacts - a system where LLMs can build full-scale apps with Next.js, Streamlit, Gradio, and other frameworks in our sandboxes. They can then publish these applications publicly and host them on E2B. In the future, E2B will be a platform where AI agents and LLMs can not only run code but also deploy it, and manage the deployed services.

We’re currently hiring across the board, so if you’re interested in helping us build this vision, join us!