E2B - AI Tools Nav

What it is for

E2B is an AI-agent sandbox cloud: your agent creates an isolated VM, runs a command or a Jupyter kernel, then pauses or kills it. Use it when the job is run the model’s code somewhere that is not your laptop, not when you need a training GPU box.

The SDK is open source (e2b-dev/E2B). Hosted Hobby is $0 plus a one-time $100 usage credit. Pro is $150/month plus per-second compute.

What makes it unique: the cover is E2B’s official plan cards — Hobby Free + $100 credits, Pro $150/month, concurrent-sandbox and session-length rows on the same grid.

Key features

Sandboxes — isolated cloud VMs your agent starts in hundreds of milliseconds and talks to over the SDK (Pricing; Billing).
Commands + files — sandbox.commands.run(...) and a filesystem API so the agent can exec, read, and write without SSH (Intro video is first-party adjacent; official Build System video below).
Pause / resume — sbx.pause() keeps filesystem state without a running VM; paused sandboxes have no TTL on the docs (Billing).
Templates (Build System 2.0) — describe the image in code so the next sandbox boots with deps already installed (Build System 2.0).
Code Interpreter path — a Jupyter-shaped runtime when the job is “run this Python and return the chart” (e2b.dev).
Lifecycle webhooks — sandbox.lifecycle.killed (and friends) so you can close the loop when a box dies (Calculate a sandbox).

Who it's for

Best fit: agent authors who will create/kill sandboxes in code and watch per-second CPU + RAM.

Poor fit: teams that want a published H100 VM they SSH into (Lambda).

Pricing

Verified 2026-10-22 (Asia/Shanghai) from e2b.dev/pricing and Billing & limits. You pay while a sandbox is running. Paused / killed boxes do not burn compute.

Hobby — $0/month. One-time $100 usage credit. 20 concurrent sandboxes. 1 hour max continuous runtime. 10 GiB disk included. No card.
Pro — $150/month + usage. 100 concurrent (add-ons to 600 / 1,100). 24 hour continuous runtime. 20 GiB disk. Custom vCPU / RAM.
Enterprise — custom. Docs list a $3,000/month minimum on the estimator. BYOC / multi-day sessions.

Published usage rates — vCPU $0.000014/s ($0.0504/vCPU-hour); RAM $0.0000045/GiB-s ($0.0162/GiB-hour). Default box: 2 vCPU + 4 GiB (pricing page) or 2 vCPU + 512 MiB (FAQ example) — copy the size you will create. Disk is included. No GPU sandboxes on the public card.

Limitations / costs

Hobby is 1 hour continuous and 20 concurrent. Pause to reset the clock; do not treat it as a long job.
$150 does not add credits. It only raises limits; usage still meters.
A $0 credit balance blocks new sandboxes if no payment method is on file.
Not a GPU marketplace. For H100 rows, see Runpod.

Alternatives

Daytona — sandboxes with a published GPU menu if the agent needs an H100.
Modal — functions if you do not want a VM-shaped sandbox.
Runpod — per-second GPU Pods if the job is train, not exec.

Official tutorial

Official E2B channel video: E2B Launches Build System 2.0 (oEmbed author: E2B | The AI Agent Cloud; CEO Vasek).

Written path: Pricing plus Billing.

Website

https://e2b.dev

Repository

https://github.com/e2b-dev/E2B

Source

Catalog row curated for AI Tools Nav. Cover is E2B’s official Hobby / Pro $150 cards. Tutorial: Build System 2.0 on @e2b-dev (oEmbed).

Last verified

2026-10-22 (Asia/Shanghai) editorial check.