This is a short note on something we feel strongly about.
Most AI products today work by sending your data to the vendor. Your emails, your client conversations, your financials, your contracts. All of it leaves your business and lands on a server somewhere else, where the vendor processes it and sends back an answer.
This is fine for some use cases. Asking a chatbot to summarise a public article. Generating a logo. Translating a paragraph.
It is not fine for a digital employee that runs your business.
The problem with the standard model.
When your data leaves your server, you lose three things at once.
- Privacy. The vendor sees everything. Their staff, their auditors, anyone they share access with.
- Control. If the vendor changes their terms, you have already given them your data. You cannot un-share it.
- Continuity. If the vendor goes down, gets acquired, or goes out of business, your operations stop.
How we do it.
Every digital employee we build runs on your server. Not ours. Your infrastructure, your data, your keys. We use frontier AI models through APIs, but those APIs are stateless. The model sees the request, gives the answer, and forgets. Nothing persists outside your environment.
If your vendor sees your data, your data is no longer yours.
The model providers we use have terms that explicitly prohibit training on API traffic and prohibit retention beyond the immediate response. We pick providers that hold to that, and we layer our own controls on top.
The trade-off.
This approach is more work to set up than the standard SaaS model. You need real infrastructure. We help with that. You also need to actually own what gets built, which means we transfer the system to you at the end. Most vendors will not do that, because their business depends on you not being able to leave.
Ours does. That is the whole point.
