Who it's for
- Teams whose files cannot go to a cloud AI service: client assets, medical, legal, unreleased work.
- Studios paying for several AI subscriptions that would rather own the capability.
- Anyone who wants to know what a small budget can actually run before buying hardware.
What you get
- A complete studio on one machine. The AI Studio experiment turned a $500 secondhand PC into a headless GPU server running 25+ open models through one web app, reachable from any device.
- Infrastructure that heals itself. Self-healing services, no cloud, no subscriptions, and discipline about what fits in 8GB of video memory.
- Local models that admit what they don't know. dnswatch names the company behind every network request with a model on my own GPU, and marks each answer as verified fact or model guess.
Open models
25+
on one $500 box
Monthly cost
$0
no cloud, no subscriptions
Companies contacted
154
by the devices in one home, found in one afternoon
Requests blocked
1 in 5
after dnswatch went live
How it works
- List what you need to generate and what must never leave your network.
- Match models to the hardware you have, or to the smallest budget that runs them.
- Set it up behind one interface your team can use from any device.
- Make it recover on its own and show you when it is guessing.
Questions
What hardware do we need?
Less than you might think. The AI Studio experiment runs 25+ models on a $500 secondhand PC with an 8GB graphics card. The limits are real, and the experiment documents them.
Is the quality as good as cloud tools?
For many tasks it is close, for some it is not. The point of a test first is to find out which on your own files before you commit.
Can it run models in Hebrew?
Speech and text models handle Hebrew at different levels, so I test on Hebrew inputs before recommending one.
See the work. The AI Studio experiment · dnswatch
Want to talk about your team? A two-line email is enough: ilan@ilans.net.