Deploy code,
run sandboxes, pay per second.
Spin up persistent sandbox environments and deploy full projects — without an account, without infrastructure configuration, just API calls. Interact via process, filesystem, and network APIs. Billed per second of active runtime.
Built for agents that need real compute behind them.
Four patterns, each relying on a different property of persistent sandboxes — short-lived isolation, stateful build pipelines, running services, or tiered compute.
Run code in isolated, persistent environments.
Agents that generate, test, or execute code need a safe place to run it. Spin up a sandbox per session, interact via the process API, read results back over the filesystem or network API, and tear it down when the task is done. Per-second billing means a five-second script costs five seconds — not a full VM-hour.
Run a full build inside one environment.
A build agent checks out code, installs dependencies, runs tests, and produces an artifact — all inside one sandbox that stays alive through every step. No reinstalling packages between stages, no shared filesystem collisions with parallel runs. Tear it down when the build finishes; spin up a fresh one for the next. Pick a tier sized to the build, not to peak monthly demand.
Deploy and expose a service in one flow.
Upload a tarball or JSON file payload to a sandbox, and the gateway handles dependency installation and process startup automatically. Once the service is running, create a preview URL to expose it publicly — no DNS configuration, no reverse proxy to manage. The entire flow is API calls.
Keep context alive across agent steps.
Agents running multi-step tasks — data processing pipelines, build jobs, long evaluations — benefit from an environment that persists between steps. Intermediate results stay on the filesystem; processes started in one API call are still running in the next. No re-setup cost between steps.
Real compute, without the setup.
Sandboxes are persistent environments you interact with via process, filesystem, and network APIs. Deploy code by uploading files — JSON or tarball — and the gateway handles the rest: dependency installation, process startup, preview URL creation. Choose a memory tier that matches your workload; billing runs per second of active runtime.
- Persistent sandboxes
- Per-second billing
- No account setup
- Preview URLs included
The honest answers.
If something below doesn't cover your case, ping us — we answer directly, no SDR funnel.
What is a persistent sandbox?
+
A sandbox is a running environment that stays alive between API calls. Unlike ephemeral execution environments that reset on every request, a persistent sandbox keeps its process state, installed packages, and filesystem intact across multiple interactions. You create it once, use it repeatedly, and tear it down when you're done.
Do I need to create an account?
+
No. You call the API; the gateway provisions and manages the underlying infrastructure entirely. No account to create, no API key to manage on your side.
How does file deployment work?
+
Upload your project files as a JSON payload or a tarball to the deploy endpoint. The gateway extracts the archive, installs dependencies, and starts the process automatically. Once running, you can create a preview URL to expose the service publicly.
What APIs can I use to interact with a sandbox?
+
Three APIs: process (run commands, read stdout/stderr, send input), filesystem (read and write files, list directories), and network (create preview URLs, inspect open ports). All three are available from the moment the sandbox is created.
How do tiers work?
+
Tiers map to memory and CPU allocations. Choose a tier when creating a sandbox based on your workload's requirements. Billing runs at the tier's per-second rate for the duration the sandbox is active. You can create sandboxes at different tiers for different tasks.
What happens when I tear down a sandbox?
+
The environment is stopped and billing ends immediately. Filesystem contents are not persisted after teardown — if you need to retain output, read it via the filesystem API before tearing down.