Say you ask an AI agent to compare two stock exports and produce a spreadsheet of the differences. The model can write a script. To finish the job, the agent needs somewhere to open the files, run the script and save the result. That workspace is the product behind E2B, a Czech-founded startup that began in Prague and now operates from San Francisco. E2B’s company announcement
On 5 November 2026, E2B announced that its annualized run rate had reached US$120 million in October. It also said the platform had reached two billion sandbox starts by October, up from one billion in June. The figures give a sense of the scale of a product that many end users never see. E2B’s announcement · Lupa.cz coverage
What E2B gives an AI agent
A sandbox is a separate computing environment. In E2B’s case, each one runs inside its own small virtual machine. Depending on how the application is set up, the agent can use it to execute code, work with files or use a browser. The application can define what the environment has access to, including which network connections it allows. E2B product documentation · E2B’s explanation of isolation and network controls
Gumloop, an automation platform, offers a concrete example. According to E2B’s customer case study, when a Gumloop agent needs to execute code, it runs the code in an E2B sandbox and returns the result to the conversation. Gumloop also uses that infrastructure for custom workflow steps that users describe in plain language. The customer interacts with Gumloop while E2B runs the code behind the scenes. Gumloop case study
From Prague to San Francisco
Co-founders Václav Mlejnský and Tomáš Valenta started E2B in a Prague apartment, according to the company. They moved to San Francisco after an early demo gained attention in 2023. In July 2025, E2B raised a US$21 million Series A led by Insight Partners. The founders took a problem they had started working on in Czechia and built a product that developers elsewhere could try and integrate themselves. E2B’s founding story and November announcement · Insight Partners’ Series A announcement
What the figures tell us
An annualized run rate projects the current pace of revenue across a year. E2B’s release gives the US$120 million figure in its headline and describes the milestone as above US$100 million in the body. The release gives no breakdown of how the run rate was calculated. Sandbox starts are a cumulative activity count. One customer can start many environments, and a start alone tells us little about whether the task succeeded. E2B’s announcement · E2B’s company timeline
How we see it at Kodo
We like that E2B grew out of such a straightforward problem. Once an agent needs to do something with a file or a piece of code, somebody has to give it a place to work. E2B has made that part available to other teams, and Gumloop’s example shows how it fits into a product people already use.
When we design an automation, we care about where the work happens and who can check the result. For the stock example, we’d start with copies of the two exports and let the agent prepare a list of discrepancies for someone to review. That’s a small, useful job on its own. It also gives the business a way to judge the output before letting the automation do more.
Kodo helps companies turn recurring work into practical automations, from getting data out of existing systems to deciding where someone should review the result. Contact Kodo
Related reading
- Duvo raises $32M for enterprise AI automation looks at another Czech-founded company working on how AI agents fit into business operations.
- Czech startups: sectors, funding and opportunities gives a broader view of the Czech startup ecosystem.