E2B

E2B is an open-source runtime and cloud service for running AI-generated code safely. Each sandbox is a Linux VM built on Firecracker microVMs, which reportedly starts in about 150ms. Agents use it to execute code, process data, run tools and control desktop environments. Developers create sandboxes through Python and JavaScript/TypeScript SDKs or a CLI. A Code Interpreter SDK keeps state between code runs and returns results such as charts. Templates define the starting environment, and sandboxes can be paused and resumed with their filesystem and memory preserved. The platform suits AI product teams, coding-agent builders and enterprises that need hardware-level isolation. Reported users include Perplexity, Hugging Face, Groq and Manus. Pricing combines a plan fee with per-second usage. The free Hobby plan comes with a one-time $100 credit, 10 concurrent sandboxes and sessions of up to one hour. Pro costs $150 a month plus usage and allows 100 concurrent sandboxes and sessions of up to 24 hours. Concurrency add-ons raise that limit to 600 or 1,100. Enterprise starts at a $3,000 monthly minimum and offers bring-your-own-cloud deployment. Usage is billed at $0.000014 per vCPU-second and $0.0000045 per GiB-second of RAM, and disk is free. The main limitation is that sandboxes are CPU-only, with at most 8 vCPUs and 8 GiB RAM. Reviewers also say costs are hard to predict because billing runs on wall-clock time, and some report immature persistence and slow pausing.