Run anything.Share nothing.
A microVM of its own for every run. Started with one API call, billed by the second.
Free for 7 days on every plan. Cancel anytime.
7c1e9a40 python:3.13-slim · 4 vCPU · 8 GB
$ POST /v1/sandboxes {"image": "python:3.13-slim", "resources": {"vcpu": 4, "memory_mb": 8192}}
201 7c1e9a40 running ready in 0.40 s
$ POST /v1/sandboxes/7c1e9a40/files?path=agent/plan.py
201 2.4 KB
$ POST /v1/sandboxes/7c1e9a40/executions {"argv": ["python", "agent/plan.py"]}
step 1/3 read the issue
step 2/3 write the fix to app/checkout.py
step 3/3 run the tests: 14 passed in 2.1 s
exit 0 3.42 s
$ DELETE /v1/sandboxes/7c1e9a40
stopped microVM destroyed
Ready in 0.40 s
Uptime 00:00:00
- 0.4 s median boot
- One microVM per sandbox
- $0.0162 an hour, small
- Python, Node and Alpine images
Boot time is the median of 100 boots on our production KVM host (p95 0.41 s).
Isolate. Six layers between your code and everyone else.
Run. Commands, files and output, live.
- stdout and stderr stream as they print
- Upload inputs, download results
- Cancel anything that hangs
Sample output
$ python -c "import sys; print(6 * 7); print('warn', file=sys.stderr)"
42
stderr warn
exit 0 182 ms
$ sleep 600
cancelled after 4.0 s
Pay. By the second, from a plan that comes back as usage.
- Small, default and large, priced per vCPU and GiB
- Usage past your plan comes from a dollar wallet
- A sandbox stops on its own when its time is up
Sample: cost so far, by the second
30:00- small2 vCPU · 4 GB · $0.0162/h$0.0081
- default4 vCPU · 8 GB · $0.0324/h$0.0162
- large8 vCPU · 16 GB · $0.0648/h$0.0324
Every run gets its own kernel, disk and network. Nothing is shared, and nothing is left behind.
Three calls from nothing to output
Example requests
curl -X POST https://api.sandlayer.dev/v1/sandboxes \
-H "Authorization: Bearer $SANDLAYER_API_KEY" \
-H "Idempotency-Key: $(uuidgen)" \
-H "content-type: application/json" \
-d '{"image":"python:3.13-slim"}'Plans that come back as usage
Every plan dollar is usage, billed by the second, and every plan is free for its first 7 days.
Starter
For one developer or a small agent fleet.
$18per seat / month
- $18 of usage a month
- 100 sandboxes at once
- 12 starts a minute
Pro
For products that run sandboxes all day.
$90per seat / month
- $90 of usage a month
- 300 sandboxes at once
- 30 starts a minute
Scale
For platforms that hand sandboxes to their own users.
$450per seat / month
- $450 of usage a month
- 1,000 sandboxes at once
- 65 starts a minute
Enterprise
For the largest fleets.
$1,800per seat / month
- $1,800 of usage a month
- 2,000 sandboxes at once
- 90 starts a minute
Questions, answered
How are sandboxes isolated?
Each sandbox is its own Firecracker microVM with its own kernel, disk and network interface, run by an unprivileged machine monitor. Sandboxes never share a container or a guest with each other.
How long does a sandbox live?
Until the lifetime you gave it runs out, at most an hour, or until you stop it, whichever comes first.
What happens to my files when it stops?
They are gone: stopping destroys the microVM with everything in it. Download what you need before you stop it.
Can a sandbox reach the internet?
Yes, unless you create it with no network. Traffic leaves through Cloudflare WARP rather than from our servers’ addresses, and mail ports are blocked.
How am I charged?
By the second, for the vCPU and memory a sandbox holds while it runs: from the usage your plan includes, then from your wallet. The console shows what each sandbox has cost so far.
Is there an interactive shell?
Yes. The console opens a real terminal in a running sandbox, with Ctrl+C, REPLs and full-screen programs, beside its commands and file transfers. The API opens one as a WebSocket.
Give your next run a machine of its own
Make a key, send one request, and watch it run.
Free for 7 days on every plan. Cancel anytime.