Agent sandboxes.Up to 6× cheaper.
A full Linux microVM per agent, with its own kernel.
Start free, no cardIntegrations
Run any code.
A Linux machine for each agent.
1from withruntime import Sandbox 2 3def solve(task, model): 4 with Sandbox.create(vcpu=2, memory_mib=4096) as sbx: 5 sbx.exec("git clone https://github.com/acme/api") 6 sbx.exec("cd api && npm ci") 7 8 for step in range(40): 9 plan = model.next_step(task) 10 if plan.done: 11 break 12 sbx.files.write(plan.path, plan.code) 13 run = sbx.exec(plan.command) 14 model.observe(run.exit_code, run.stdout) 15 16 tests = sbx.exec("cd api && npm test") 17 return tests.exit_code == 0
- Pauses when idlePays storage only, and wakes on the next request.
- No time limitRuns as long as the work does.
- Root and Dockerapt, sudo and Compose inside.
Bring any environment.
Your own Dockerfile.
- Any Dockerfile or registry image
- Built once, versioned
- Docker runs inside
1FROM python:3.12-slim 2RUN apt-get update && apt-get install -y ffmpeg 3RUN pip install --no-cache-dir torch playwright 4RUN playwright install --with-deps chromium 5COPY . /app 6WORKDIR /app
$ runtime image build . -t web:v1 ✔ web:v1 $ runtime sandbox create --image web:v1 ✔ sbx-7f3a running
web:v1every sandbox starts from it
Drive a real browser.
Click, type, screenshot.
Order confirmed
Trail Runner 2, size 42 · order A-1042
Pause it. Fork it. Snapshot it.
Plug it in. All from the same sandbox.
Everything you need, built in.
Secure by design.
Every sandbox is its own machine.
- Its own machine Each sandbox is a virtual machine with its own kernel, disk and memory.
- Secrets stay outside A sandbox holds a placeholder; the real value goes in on the way out.
- Public internet only Private and internal addresses are refused, and any sandbox can be cut off at once.
- Encrypted backups Copies of your volumes and snapshots are encrypted with your organization's own key before they leave the server.
- Audited data centers SOC 2 Type 2 and ISO 27001, in Virginia, United States.
- Sign-in you control Single sign-on, two-step sign-in, four roles and an audit log, free.
Switch by changing
one import.
- Your codewritten for E2B, Daytona, Vercel Sandbox, Blaxel
from e2b import Sandbox
- One importthe only line that changes
- from e2b import Sandbox+ from withruntime.e2b import Sandbox
- InstallPython or JavaScript
$ pip install withruntime
- Save moneyfor the same 1,000 sandbox hours
up to 5× less
Questions
before you switch.
Why is it so much cheaper?
E2B and Daytona bill every vCPU a sandbox holds for as long as it runs. Runtime bills the CPU your code uses, $0.025 per active vCPU-hour above a default floor of a twentieth of a vCPU, and $0.0075 per GiB-hour of memory, the lowest memory rate of the providers we compare that price memory on its own. There is no plan fee. Time spent waiting on a model costs only memory and that floor, so the same agent job costs 42% to 88% less than at fourteen other providers. See the pricing.
What does an idle sandbox cost?
Only its paused storage. A sandbox pauses itself after 300 seconds with nothing happening in it: no request, no running command, no open connection, no traffic and no CPU use. It keeps its memory, files and processes for $0.08 per GB a month, and the next request wakes it. See idle and paused costs.
Do I have to rewrite my code?
No. Code written for E2B, Daytona, Vercel Sandbox or Blaxel runs by changing one import, in JavaScript and Python. Or paste the agent prompt into your coding agent: it switches the project on a branch, tests it on the free usage and tells you what you save. Read the switch guide.
What is included for free?
100 free hours of a 1 GB machine: 100 GiB-hours of memory and 25 vCPU-hours of active CPU, no credit card, once across your first 30 days, then every month renewed on the 1st once you have paid. Without credit, up to four sandboxes at once with 2 vCPUs and 4 GiB each. Past it, prepaid credit from $10, with no plan fee, and credit you add within 30 days of signing up is matched, up to $1,000 in total. See what is included.
How is each sandbox isolated?
Every sandbox is a Firecracker microVM with its own Linux kernel, disk and guest environment, on dedicated servers Runtime operates. CPU, memory, network rules and time limits are enforced on the host, outside the guest. Read the security model.
Is there an uptime promise?
Yes. Paid accounts are promised 99% API uptime every calendar month, measured from outside our servers and published on the status page. If a month falls short, 10% of that month's charges comes back as credit automatically, with no claim to file. Read the Uptime Promise.
How big, and for how long?
A paid sandbox takes up to 16 vCPUs and 64 GiB of memory, and an account runs up to 100 at once from its first top-up. A sandbox runs while it works and pauses itself when idle; there is no time limit unless you set one. See the sizes.
What is Runtime Cloud?
Runtime Cloud is a cloud for AI agents' code: Linux sandboxes, each a Firecracker microVM, billed on the CPU they use. About Runtime.
100 free hoursin your first 30 days.
of a 1 GB machine
- No credit card
- Four sandboxes at once, 2 vCPU and 4 GiB each
- Then prepaid credit from $10, no plan fee