- Only pay for the portion of CPU, memory, and disk you actually use within a sandbox. Agents spend most of their time blocked on I/O, so observed-usage billing costs a fraction of reserved-vCPU pricing (see the comparison below).
- Access to low-latency local NVMe disks.
- Run indefinitely without a fixed maximum lifetime.
- Pause and resume sandbox state, including open network connections, for any duration. In-flight work survives the pause with no save/restore code on your side.
- Automatically sleep Sailboxes while waiting on Sail inference calls.
Sailboxes are currently in beta. APIs and operational behavior may change as
we stabilize the product.
Provider comparison
| Provider | vCPU price | Sleep during inference | Max runtime | Memory snapshots | Network snapshots | Local NVMe disks | Docker-in-Docker | Start/resume time |
|---|---|---|---|---|---|---|---|---|
| Sailboxes | $0.03/actively-used vCPU-hour | Yes | No fixed limit | Yes | Yes, with packets cached for 10 minutes | Yes | Yes | <2s |
| Modal | $0.071/reserved-vCPU-hour | No | 24 hours | Alpha | No | No | Alpha | <500ms |
| E2B | $0.0504/reserved-vCPU-hour | No | 24 hours | Yes | No | No | Yes | <1s |
| Vercel Sandbox | $0.128/reserved-vCPU-hour | No | 5 hours | No | No | Yes | Yes | <1s |
| Daytona | $0.0504/reserved-vCPU-hour | No | No fixed limit | Experimental | No | No | Yes | <500ms |
Start here
- Quickstart: create a Sailbox, run commands, expose a service, and clean up.
- Pricing: understand observed-usage billing dimensions and per-hour rates.
- Images: choose a base image and build custom images with packages, commands, environment variables, and local files.
- Networking: expose HTTP services, raw TCP ports, and SSH from a Sailbox.
- Filesystem: read and write files at runtime, stream large files, and decide what belongs in an image.
- Lifecycle: checkpoint, fork, pause, sleep, resume, upgrade, and terminate Sailboxes.