E2B: Secure Sandboxes for AI Agents, Explained
E2B gives AI agents their own computers to work in. When an agent writes code, runs shell commands, installs packages, analyzes files or operates a desktop app, it does it inside an E2B sandbox, a small, isolated virtual machine that boots from a snapshot, rather than on the company’s own servers. That isolation matters because AI-generated code is unpredictable. E2B calls itself “The AI Agent Cloud,” and says more than 100,000 teams have run a sandbox, 94 of the Fortune 100 have signed up, and its sandboxes passed two billion starts in October 2026. Its runtime is open source under the Apache 2.0 license.
The quick answer: E2B is one of the most established sandbox platforms for AI agents, used by companies like Manus, Genspark, Hugging Face and Groq. It’s the default choice for running AI-generated code safely at scale, and it also offers full desktops for computer-use agents. Enterprises can run it inside their own cloud account. Everything below comes from E2B’s site, pricing, developer summary, about, security and customer pages.
E2B at a glance
Checked against E2B’s site, pricing page, developer summary and about, security and customer pages on November 13, 2026.
| E2B | |
|---|---|
| What it is | Isolated cloud sandboxes where AI agents run code, use tools and operate desktops |
| Isolation | A Firecracker microVM with its own kernel for every sandbox |
| Capabilities | Shell, filesystem, network, storage, browser, desktop, pause and fork |
| Works with | Any model or framework, including the OpenAI Agents SDK, Claude Code and Codex |
| Open source | Apache 2.0 runtime and SDKs, about 14,800 GitHub stars |
| Pricing | Free Hobby plan with $100 in credit; Pro $150 a month plus usage |
| Company | FoundryLabs, Inc., San Francisco and Prague; founded 2023 by Vasek Mlejnsky and Tomas Valenta |
What E2B does
E2B’s about page explains why agents need a different kind of computer: traditional containers and virtual machines “were designed to run software defined ahead of time,” while agents generate code and choose actions as they go. They need an environment that’s isolated from the host, fast enough for interactive products, stateful when work takes time, and equipped with the tools a person uses.
E2B’s answer is “a Firecracker microVM with its own kernel” for every agent session. Developers create one with a line of code, and the agent gets a shell, a filesystem, network access, storage, a browser and, optionally, a full desktop. Sandboxes can be paused and resumed, forked, given secrets scoped to that sandbox, and exposed through public URLs, so an app the agent builds can be previewed live. Network traffic follows policies you set.
What teams use it for
E2B’s homepage and customer page describe several patterns:
- General-purpose agents that need a full virtual computer with tools, files, a browser and long-running state. Manus uses E2B to give its agents virtual computers, and Genspark runs a general-purpose agent for millions of users on it.
- App generation, where a prompt spins up coding agents that build an app and serve it live, as on the vibe-coding platform Anything.
- Data analysis and code actions, running Python and JavaScript with files, charts and tables.
- Computer use, through Desktop Sandbox, a graphical Linux environment that an agent drives with clicks and typing, streamed live.
- Training and evaluation, where thousands of identical machines run, grade and clear in parallel. Hugging Face used parallel E2B sandboxes to reproduce DeepSeek-R1 in its Open R1 project, and LM Arena built web-development evaluations on it.
Other named customers include Groq, Rogo and StackAI.
Security and deployment
Security is central to E2B’s pitch. Its security page says isolation happens “at the hypervisor boundary, not the container or process boundary, which is what makes it safe to run untrusted or AI-generated code.” E2B is SOC 2 Type II compliant and signs HIPAA business associate agreements on Enterprise plans.
Enterprises can use E2B’s cloud, or deploy sandboxes into their own AWS, Google Cloud or Azure account through E2B’s bring-your-own-cloud option, where data stays inside the customer’s network and E2B operates the cluster. Because the runtime is open source, teams can also operate it themselves.
Developer experience
E2B has Python and JavaScript SDKs, a CLI, a REST API and an MCP gateway that connects more than 200 MCP servers to sandboxes. Code Interpreter is a ready-made sandbox for running code and returning rich results. E2B also publishes instructions for AI agents to set themselves up, a public MCP server for its documentation, and a cookbook of examples.
Pricing
E2B’s pricing charges for each second a sandbox runs, on top of a plan:
| Plan | Price | Includes |
|---|---|---|
| Hobby | Free | $100 one-time usage credit, sessions up to 1 hour, 10 concurrent sandboxes, no card required |
| Pro | $150 a month plus usage | Sessions up to 24 hours, 100 concurrent sandboxes (expandable to 1,100), custom CPU and RAM |
| Enterprise | Custom | Multi-day and persistent sessions, tens of thousands of concurrent sandboxes, custom sizes, bring your own cloud |
Usage is billed per second by CPU and memory. The default sandbox, 2 vCPUs and 4 GB of RAM, costs about $0.000046 a second, or roughly $0.17 an hour. E2B also offers a one-time $20,000 in credits plus the Pro plan for eligible early-stage startups and research teams.
The company
E2B was founded in 2023 by Vasek Mlejnsky and Tomas Valenta and operates from San Francisco and Prague under the legal name FoundryLabs, Inc. According to its about page, the founders started by connecting an agent to sandbox technology from earlier developer-tool work, then shifted focus from the agent to “the reusable machine underneath it.” E2B shipped SDK 1.0 and Desktop Sandbox in 2024, announced a $21 million Series A in 2025, and passed one billion sandbox starts in June 2026 and two billion that October.
How E2B compares
E2B competes with other agent sandboxes and computer-use platforms, including Daytona, Cua, Tensorlake and Orgo. Orgo focuses on persistent Linux, Windows and Mac desktops that stay on until stopped, while E2B is strongest at fast, isolated, high-volume sandboxes for code and agent sessions, with desktop support alongside. For agents that only need a browser, platforms like Browserbase and Browserless are simpler. E2B’s distinguishing traits are its scale, microVM isolation, open-source runtime, enterprise deployment options and a long list of well-known AI customers.
Agents need sandboxes when they write and run their own code. For day-to-day work in the tools you already use, Carly connects to thousands of apps directly, so from ChatGPT, Claude or a text message it can sort email, schedule meetings and update your CRM.
Who E2B suits
E2B is a good fit if:
- Your agent writes and runs code, and you need to isolate it safely.
- You run many sandboxes at once, for a product, evaluations or model training.
- You need computer-use desktops alongside code execution.
- You need SOC 2, HIPAA or deployment inside your own cloud.
Look elsewhere if:
- You need Windows or Mac desktops that stay on for weeks.
- Your agent only needs a browser.
- You want a finished assistant rather than infrastructure.
FAQ
What is E2B?
E2B is a cloud platform that gives AI agents isolated sandboxes, each a Firecracker microVM, to run code, use a shell and files, browse and operate a desktop safely.
Is E2B open source?
Yes. E2B’s runtime and SDKs are open source under the Apache 2.0 license, and teams can use E2B’s cloud, deploy into their own cloud account or run the runtime themselves.
How much does E2B cost?
E2B’s free Hobby plan includes a $100 one-time credit. The Pro plan costs $150 a month plus usage, billed per second by CPU and memory; a default sandbox costs roughly $0.17 an hour.
Does E2B support computer use?
Yes. E2B’s Desktop Sandbox provides a graphical Linux desktop that an agent controls with mouse and keyboard actions, streamed live.
Who founded E2B?
Vasek Mlejnsky and Tomas Valenta founded E2B in 2023. It announced a $21 million Series A in 2025.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
