E2B
Open-source cloud sandboxes for AI agents: Firecracker microVMs that run code, shells and desktops, with Python and JS SDKs and per-second pricing
Overview
E2B gives AI agents their own isolated cloud computers. Each sandbox is a Linux Firecracker microVM that an agent can use to run code, shell commands, a browser or a full graphical desktop, controlled through Python and JavaScript SDKs. Add-on packages cover a code interpreter with charts and tables, and a desktop sandbox for computer-use agents. Sandboxes can be paused and resumed from memory or file-system snapshots, forked into up to 100 copies, given custom templates, and locked down with egress rules and secrets injected at the proxy. You can use E2B Cloud (US, EU and APAC regions), self-host the Apache-2.0 E2B Embed package on Linux with KVM, or run the data plane in your own AWS, GCP or Azure account on Enterprise. Founded by Vasek Mlejnsky and Tomas Valenta, E2B raised a $21M Series A led by Insight Partners in July 2025.
The Verdict
Who Should Use E2B?
Best For
- Teams building agents that must run untrusted, AI-written code safely
- Data-analysis and code-interpreter features inside AI apps
- Computer-use agents that need a Linux desktop
- Companies that want an open-source option they can self-host
Not Ideal For
- GPU workloads: E2B sandboxes are CPU-only
- Very long-running jobs on the free plan (one-hour session cap)
- Teams that need a full dev-environment product with IDE and Git tooling
What's Great
- Hardware-level isolation with one Firecracker microVM per sandbox
- Apache-2.0 SDKs, plus a self-hostable single-node package
- Public per-second pricing with $100 of free usage
- Snapshots, pause and resume, and forking for agent workflows
- Used by Hugging Face, Perplexity, Groq, LMArena and Manus
Watch Out For
- Session length caps: 1 hour on Hobby, 24 hours on Pro
- No GPU sandboxes
- Pro is $150 a month before usage, a jump from the free plan
- No official cold-start figure is published; third-party numbers vary
- Most adoption claims come from E2B's own marketing
Pricing
Usage is billed per second: $0.000014 per vCPU-second and $0.0000045 per GiB of RAM per second. Disk is free. Sandboxes range from 1 to 8 vCPU and 1 to 8 GiB RAM.
View all features & details
Key Features
- Python and JavaScript SDKs
- Code interpreter package with rich outputs
- Desktop sandbox for computer-use agents
- Custom templates, snapshots, pause and resume, forking
- Egress control by CIDR or hostname; secrets at the proxy
- HTTPS port exposure, PTY, streaming output, webhooks
Use Cases
- Running code generated by LLMs
- Data analysis and charting in chat apps
- Coding and computer-use agents
Deployment
- E2B Cloud in US, EU and APAC; SOC 2 Type II
- E2B Embed: self-host with Docker Compose, Terraform or Kubernetes
- BYOC data plane on Enterprise
How It Compares
| Feature | E2B | Daytona | Modal |
|---|---|---|---|
| Isolation | Firecracker microVM | See vendor | See vendor |
| Open source | Yes (Apache 2.0) | Repo frozen, closed source since June 2026 | No |
| GPUs | No | Preemptible GPUs | Yes |
| Free credit | $100 | $200 | See vendor |
| vCPU price | $0.000014 per second | $0.0504 per hour (about the same) | See vendor |
Sources: E2B pricing · E2B on GitHub · Daytona pricing · LogRocket comparison
Is this your tool? Add a badge