Naïve agent computers now run on Tenki Sandboxes
Naïve's AI employees incorporate companies, launch campaigns and write code. Every task they run now gets its own fresh Tenki sandbox.
Guzman Pintos
2 min read
Starting today, every Naïve agent computer runs on Tenki Sandboxes.
Naïve builds infrastructure for agents that operate businesses end to end. You describe the company, and Naïve deploys a team of AI employees to run it. They incorporate it, set up banking, launch marketing campaigns, run outbound and write the code. Each agent works under its own email address, phone number, domains and third-party accounts.
Why every task gets its own VM
Every one of those actions needs a real machine underneath it, isolated from everything else the agent touches.
That matters more when the agent isn't just writing code. An agent that holds a spend-capped card, sends email from the company's domain and pushes to production shouldn't inherit files, processes or credentials from the task before it. It needs a clean environment every time.
So each task a Naïve agent runs boots a fresh Tenki sandbox: a Linux VM on Tenki's own bare-metal servers, interactive in under a second and destroyed when the job is done.
What Naïve users get
Naïve says that, on average, its users will see:
- about 30% lower sandbox costs
- 72.3% lower sandbox latency
Two properties of Tenki Sandboxes matter for a workload like this. A sandbox's disk image already sits in Tenki's storage cluster, so starting one means mounting that disk and reserving cores rather than copying an image to a machine. And billing is per second, so a task that finishes in forty seconds pays for forty seconds.
Run your own agents on Tenki
Tenki Sandboxes are available to any team building agents. The quickstart gets you from install to a running VM, and if your agents call APIs with real credentials, Tenki Secrets keep the tokens outside the sandbox.
Thanks to Sean Dorje, Dennis Zax and the Naïve team for building on us.
