Offloads 80% of prompts (non-frontier)
Your AI tools send Vault the work that doesn’t need a frontier model. It runs locally, so those tokens stop costing you.

Private AI box · Launching Feb 2027
Vault is a private AI box that runs open-source models on your desk. Unlimited tokens, no per-task fees, and your data stays home.
The problem
Triage, follow-ups, research, summaries, drafts. Your agents do this all day, and you pay frontier prices for every token.
~80%: Vault estimate. Your mix may differ.
The product
Vault sits on your desk and works in the background, around the clock.
Your AI tools send Vault the work that doesn’t need a frontier model. It runs locally, so those tokens stop costing you.
Out of the box, Vault is an agent of its own, and it gets more capable as open models improve.
How it works
Connected over open standards
Leading open-source models · encrypted memory
The hard steps, and nothing else
Power and network. That’s the setup.
Point your AI tools and accounts at Vault.
24/7, on hardware you own.
Private by design
Data never leaves the box unless a step truly needs a bigger model.
Everything Vault learns about your work is encrypted on the device.
One flip and it’s off. No software toggle to trust.
It’s hardware you own, so local work is free and unlimited.
Better models, new skills and security fixes, delivered over the air.
Savings
Your monthly AI spend
Estimate assumes ~80% of your AI work can run locally. Your mix may differ.
Pricing
With Fleet, your Vault lends spare compute to encrypted jobs while idle, and you earn credit toward an upgrade. Your data & memory stay private.
$100 off with Fleet
$100 off with Fleet
$100 off with Fleet
Ships February 2027.
Why Vault
| Option | Unlimited tokens | Data stays home | No setup | Runs 24/7 |
|---|---|---|---|---|
| Vault | Yes | Yes | Yes | Yes |
| Cloud AI subscriptions | No | No | Yes | Yes |
| Other AI boxes | Varies | Varies | Varies | Varies |
| DIY Mac / GPU rigs | Yes | Yes | No | Yes |
FAQ
A private AI box that sits on your desk and runs open-source models locally. It handles the everyday work your AI tools do, and it runs a built-in agent for email, follow-ups and scheduling.
Because the work runs on hardware you own. There’s no per-token or per-task fee for anything Vault handles locally.
Only when a step truly needs a bigger model. Everything else stays on your desk, memory is encrypted, and a physical switch turns it off.
Fleet is an option on every Vault. When your Vault is idle, it lends spare compute to encrypted jobs, so you get $100 off (Core $299, Plus $499, Pro $799). You also earn credit toward an upgrade.
Yes. Jobs that use your idle compute are encrypted and kept separate from your data. Your memory and files stay encrypted on your box.
Leading open-source models, updated over the air as they improve, along with new skills and security fixes.
Yes. Vault connects to your tools over open standards, so they can send it the work that doesn’t need a frontier model.
February 2027. Join the waitlist to hear first.
Ships Feb 2027
Join the waitlist for early access. From $299.