How Genspark Uses E2B to Run a General-Purpose AI Agent for Millions of Users

Genspark's Super Agent orchestration handles everything from deep research and slide creation to making phone calls and analyzing data. Under the hood, Genspark uses E2B infrastructure to give its agents secure, isolated virtual machines where they can execute code, browse the web, and complete complex tasks autonomously.

‍

A General-Purpose Agent Needs a General-Purpose Runtime

Unlike single-purpose AI tools, Genspark’s Super Agent orchestration is designed to handle any task a user throws at it. A user might ask it to research a competitor, build a financial model, generate a presentation, or analyze a dataset — all in a single session.

This means the execution environment can’t be a simple code sandbox. Genspark’s agents need full virtual machines where they can:

• Execute arbitrary code — Python, JavaScript, shell scripts, whatever the task requires
• Install packages and dependencies on the fly, adapting to each task
• Run long-running workflows — research sessions that span dozens of steps, not just quick code snippets
• Access the web — browsing, scraping, and pulling live data as part of multi-step tasks
• Maintain a persistent state — keeping context, files, and intermediate results across an entire agent session

E2B’s Firecracker micro VMs provide exactly this: lightweight virtual machines that spin up in milliseconds, run securely in complete isolation, and support the full range of operations a general-purpose agent needs.

Enterprise-Ready from Day One

Genspark isn’t just a consumer product — enterprises rely on it for business-critical workflows. When a Fortune 500 company uses Genspark’s agent to analyze internal data or generate reports, the execution environment needs to meet enterprise standards. When choosing a sandbox provider, but more generally, an AI cloud to run your agents on, you need to be very honest in answering the following questions:

• Security & data isolation — Does this sandbox solution make your users more or less secure?
• Reliability — How much can our users bet on this tool?
• Scalability — Can our users really grow 100x with this sandbox provider?
• Speed — Is this tool slowing down your users? If yes, how much?

With E2B, each agent session runs in its own isolated Firecracker micro VM. No data leaks between sessions, no risk to Genspark’s core infrastructure. From a single user researching a topic to enterprise teams running hundreds of parallel agent tasks, E2B also scales with demand.

“A 30-step agent run can't start with a 60-second cold start. It has to feel instant. E2B lets us scale to thousands of concurrent sessions, and we couldn't have hit $250M ARR if five of our engineers were building a sandbox platform instead of the agent.”

Why E2B Over Building In-House

For a company growing as fast as Genspark (from zero to $250M ARR in 12 months) building and maintaining custom execution infrastructure would be a massive distraction. This kind of infrastructure typically requires 3-5 dedicated engineers and months of development.

E2B lets Genspark’s engineering team focus entirely on making the agent smarter and more capable, while E2B handles the execution layer: spinning up isolated virtual machines, managing resources, ensuring security, and scaling to millions of sessions.

What’s Next

Genspark is at the forefront of a new category: general-purpose AI agents capable of truly doing anything. As agent capabilities expand — longer sessions, more complex multi-step workflows, deeper enterprise integrations — the demands on the execution environment will only grow. E2B’s infrastructure is built for exactly this trajectory: secure, fast, scalable virtual machines that give AI agents the full computing environment they need to get real work done.

E2B is the open-source cloud infrastructure for AI agents. Give your agents secure virtual machines that spin up in milliseconds. Building something similar? Book a call with our engineers to learn more.