Run AI-written code without risking your own systems.
Litterbox runs untrusted code in a sealed box that cannot reach your network or your keys, returns the result, and deletes the box.
Book a demo
Free during beta
- to start a fresh sandbox
- < 2 s
- access to your network, files or credentials
- 0
- hard limit, so nothing runs away
- 60 s
Without Litterbox
- AI-written code runs on machines that hold your secrets
- One bad command can delete data or leak a key
- Building your own sandbox takes months
With Litterbox
- Every run is sealed off and thrown away
- One API call, no servers to manage
- Every run recorded, so failures can be replayed
Three steps. That is all.
Send
Your app sends the code and any files it needs
Run
It runs sealed: no network, 60 seconds, capped memory
Return
You get the output back and the sandbox is destroyed
Freeduring beta
Up to 10 hours of compute a month. Paid plans will be per second of use.
For teams building AI agents into their product.
| Works with | REST and gRPC APIs, TypeScript and Python clients |
|---|---|
| Hosting | Hosted by BKL in us-east-2 |
| Built with | Go, Firecracker microVMs, gRPC |
| Availability | Beta, by request |
See Litterbox pounce on your own data.
Thirty minutes, with the people who built it. Bring a real file, a real inbox or a real error log. No slides.
Book a demo