Gpu compute for AI agents
Serverless GPU on demand — run a model, batch a heavy job, tear it down. No cluster to keep alive.
What it's for
Get a GPU for one job without provisioning anything.
Run inference, batch a transform, or execute a heavy script.
Start it, run commands, check status, terminate — all from the conversation.
For work too heavy for an ordinary code sandbox.
How to ask your agent
Connect Vaaya, then just say it in plain language. For example:
❯Use Vaaya to run this inference script on a GPU and give me the output.
❯Use Vaaya to batch-process these files on a GPU and shut it down after.
What you get back
{
"sandbox_id": "sb_modal_71e2",
"status": "succeeded",
"stdout": "processed 1,240 rows in 38s"
}
Pricing
1¢ per sandbox command, plus the GPU time the job actually uses. Billed per call in cents, capped by your max, every call on the ledger. Failed calls are not billed.
Give your agents this.
Connect Vaaya once, then describe the outcome. It picks the right tool, runs it, and pays per call.
npx @vaaya/mcp install