What is Vakna

Vakna runs the code AI agents write in boxes: isolated machines you create, suspend, resume and fork from the vakna CLI.

A box is a whole machine with its own kernel, disk and network address. An agent creates one, runs a command in it and reads what it printed. Nothing it runs can reach your laptop, your other boxes or another customer's.

A box keeps its state between commands. Suspend it and its processes, memory and open ports wait, and compute stops billing. Resume it and the processes carry on from where they stopped.

A snapshot captures a box's memory and disk at one moment. Fork from it and you get many boxes in that state, started from memory rather than booted, so a test suite that needs ten minutes of setup runs fifty ways in parallel with the setup paid once. Those boxes form a task, and the task reports totals first, then the boxes that need you.

Every box starts in lockdown: it reaches only the hosts you allow. When the command in it needs one you haven't allowed, Vakna names the host and the command that allows it.

Every command ends with a receipt: what ran, how long it took, and what it cost.

Teams whose agents write and run code: coding agents running tests, research agents running experiments, and anyone who wants to run code they didn't write without trusting it.

The method page has the benchmark: the hardware, the region, the method and the repo. Numbers on these pages come from the CLI's receipts and point there.