README
The sandbox cloud for AI agents.
Small, powerful computers for AI agents. They boot in 400 ms, stop and resume at any time, and bill by the second.
npm install -g @platinum-dev/cli
pt init
pt sandbox create -t pt-baseUse cases
What runs on Platinum.
Coding agents
Build your agent cloud around any harness: Codex, Claude Code, OpenCode, Pi. Ready to scale.
Code interpreter
Run untrusted, AI-generated code. Stream the output. Throw the box away.
AI evaluations
Fork one snapshot into hundreds of identical parallel runs, each with reproducible state.
Data analysis
Point an agent at your data. Explore, compute, and chart in isolation.
Deep research agents
Long-running research over big inputs. Pause it, resume it, keep digging.
Vibe coding
The runtime for AI-generated apps. Any language, any framework. Expose a port, share the URL.
Features
01
Isolate
Every sandbox is a microVM. Own kernel, own disk, own network.
02
Pause
Suspend to disk mid-process. Resume in under a second.
03
Fork
Snapshot a running box. Fork copies on demand.
04
Boot
400 ms cold starts. From create call to running code.
05
Scale
Thousands of concurrent sandboxes. Capacity-aware placement.
Pricing
Same sandboxes.
vCPU
RAM
Disk
Concurrent sandboxes
Hours / month · each
Platinum
$111/mo
$331/mo
$332/mo
$365/mo
$476/mo
$507/mo
$644/mo
$682/mo
Platinum rates: $0.0168 / vCPU·hr · $0.0054 / GB RAM·hr · $0.026 / GB disk·mo
Published list prices · verified Jul 2026 · compute & disk billed per running hour (disk $/GB·mo prorated ÷ 730, after each vendor's free storage) · AWS = Lambda MicroVMs · Vercel & Cloudflare active-CPU at full util · Cloudflare = Containers · Fly = Sprites · base fees, egress & credits excluded
FAQ
Common questions.
What exactly is a Platinum sandbox?
A full microVM with its own kernel, disk, and network, hardware-isolated. That makes it a complete, ready-to-use VM, not a locked-down runtime: install packages, run daemons and long-lived processes, SSH in, expose ports, keep files. Anything you can do with a VM, you can do here. It boots in about 400 ms, driven from the TypeScript or Python SDK, the pt CLI, the REST API, or the dashboard.
How does billing work?
Per-second, only while a sandbox runs: $0.0168 per vCPU-hour plus $0.0054 per GB of RAM-hour. A stopped sandbox costs at most its disk ($0.026 per GB-month). Prepaid credits, no subscription. New accounts get $10 free.
How can it cost a third of the other sandbox clouds?
Platinum is Kortix Web Services, the same play Amazon ran when AWS productized the infrastructure behind Amazon.com. This is the infrastructure that runs Kortix in production today, on our own capacity-packed bare metal. We pay for it anyway, so productizing it hands you our economics instead of a hyperscaler markup.
Can I migrate from my current provider?
Yes. Coming from Daytona, the SDK is a drop-in: point it at Platinum with a Platinum API key, no code changes. From E2B, Modal, or anywhere else, any Docker image becomes a bootable template and the SDK shapes will feel familiar. We subsidize all migration costs and take over your existing contract when you move.Daytona drop-in docs →Talk to us about migrating →
How big can a sandbox be, and how many can I run?
Up to 16 vCPU, 32 GB RAM, and 100 GB disk per sandbox by default. Larger shapes are available on request. Hundreds of concurrent sandboxes per org out of the box, and more by arrangement. The fleet is capacity-aware, not artificially capped.
Do sandboxes keep their state?
Yes. Stop a sandbox and resume it later exactly where it was; snapshot a running box and fork identical copies on demand. Long-stopped sandboxes are archived to durable storage and resume from there. State is never silently deleted.
Can I run Platinum on-prem?
Yes. Platinum is available as a licensed on-prem deployment: the full platform on your hardware, set up and supported by us. It is not self-serve; talk to us and we will scope it together.Ask about on-prem →
About
Built by the Kortix team.
We ran 3+ million sandboxes on incumbent providers while building Kortix. We hit the cost ceiling, so we rebuilt the infrastructure we needed, and it runs Kortix in production today.
It's also how the price works. Platinum is Kortix Web Services. We run this infrastructure for ourselves anyway, and productizing it is what lets us charge a third of what the others do.
Get started