Private AI Gateway
Give your team cloud AI models without sending them personal or confidential data. One controlled entry point, inside your network.
Your data is already going to AI chats
People paste customer emails, contracts, and code into AI tools because it saves them hours. Blocking it moves the same habit to personal accounts. A gateway lets them keep the speed and keeps the data with you.
I'm sorry about the trouble. We've issued a full refund to the card •••• 1111; it should appear within 5–7 business days.
A confirmation is on its way to jane.cooper@acme-mail.com.
Best regards,
Support team
What you get
- Gateway in your network. Deployed on your servers or in your cloud account. One API for chat, IDE assistants, and internal apps.
- Detection tuned to your data. Rules plus a local model for personal data, financial details, credentials, and your own identifiers: customer IDs, contract numbers, and project names.
- Masking and restoration. Values are replaced before a request leaves and restored in the reply, so people still see real names and numbers.
- Routing and budgets. Send each request to a cloud or local model by policy, with budgets per team and per user.
- Policies. Decide what may leave, what is blocked, and what needs approval, per data type and per team.
- Audit log. Every request: who, when, which model, what was masked. Stored with you, exportable to your SIEM.
- Detection quality report. Measured on your own samples, so you know what the gateway catches before it goes live.
Six weeks, four steps
Smaller setups with one tool and rules-only detection fit into four weeks.
- Week 1
Discovery
Which tools people use, which data must never leave, sample collection.
- Weeks 2–3
Deploy
Gateway, detection rules, and the local model in your environment.
- Weeks 4–5
Tune and connect
Detection tuned on your samples, integrations with chat, IDE, and apps, and the policies.
- Week 6
Handover
Runbook, admin training, quality report, and the first audit review.
What we need from you
- A technical contact who can grant access and approve changes.
- A server or cloud account for the gateway. A GPU is needed only for the local detection model.
- 200–500 sample texts with the kind of sensitive data you care about. We sign an NDA before seeing them.
- The list of tools people use with AI today.
Questions
Does masking hurt answer quality?
For the tasks people use AI for (summaries, drafts, and analysis), usually not: placeholders keep the structure of the text. We measure it on your samples and show the difference before launch.
Which models can we use?
Any provider with an API, plus open-weight models running on your hardware. Apps talk to the gateway, so switching models doesn't require changing them.
Does it catch everything?
No detector catches 100%. We report measured detection on your samples, add rules for identifiers specific to your business, and you can block whole categories of data from leaving at all.
What does it cost to run?
A CPU server is enough for rules-only mode. The local detection model needs a single mid-range GPU. Model API usage is billed by the provider, and the gateway's budgets keep it under control.
We already block public AI chats. Why do we need this?
Blocking usually moves AI use to personal phones and accounts, where you see nothing. A gateway gives people an approved way to use AI and gives you control over what leaves.
Find out what leaves your company today.
On a 30-minute call we'll map which AI tools your team uses and what a gateway would cover. We reply within one business day.