Services · Private AI
Private AI for organisations whose data can't leave the building.
Banks, hospitals, law firms and public bodies want AI, but can't hand customer records, patient files or case papers to someone else's cloud. We build AI that runs inside your own cloud account or on your own hardware, under your keys.
Built for
- Banks and financial institutions
- Hospitals and health systems
- Law firms
- Research labs
- Regulators and public bodies
Self-hosted AI, in under two minutes
What runs inside your walls, how a question travels from an employee to the model and back without leaving your network, and how the hardware scales from one workstation to a rack. 1 min 47 s, with captions. Read the transcript.
Four rules for every deployment
Runs inside your walls
In your own cloud account, in the region you choose, or on servers you operate. Fully air-gapped if your policy requires it.
Your keys. Your logs.
Encryption keys stay in your key management service and audit trails stay in your systems. We never hold either.
Nothing calls home
No telemetry, no analytics, no traffic to us. Your documents are never used to train a model, ours or anyone's.
Delete means delete
Hard deletes that actually run, on a schedule you set. No hidden copies on our side, because there is no our side.
These are the rules a private deployment is built to, delivered and verified during your pilot. Our hosted products run on shared infrastructure and don't all meet them today. Sitr Shield does: it filters on your device and never contacts us.
Two ways to deploy
In your cloud account
A single-tenant installation in your own AWS, Azure or Google Cloud account. Models are served through your cloud's private endpoints, so prompts and documents never leave your account.
On your own hardware
Air-gapped, with open-weight models on GPUs you own. For policies that forbid any public cloud, or data that must stay in the country where no cloud region exists.
Which one fits depends on your rules and on what your own team can run. Open-weight models on your own hardware are a step below what a cloud's private endpoints offer today, and the hardware is yours to size and run. We say that in the first conversation, not after.
What we build
- Cognoir Enterprise
- Ask questions of your own documents and get answers cited to the exact passage, with code, not AI, checking every quote. Deployed single-tenant in your environment. cognoir.com/enterprise
- Custom AI on your data
- AI and data systems built around the work you actually do: reviewing case files, answering from clinical protocols, searching loan records. Built inside your infrastructure from the first line.
- Compliance-grade systems
- Software built to a regulation's letter: insert-only records, access logs on every lookup, data-subject requests and deletion that runs on schedule.
- Data pipelines and analysis
- Collecting, grouping and scoring large volumes of sources on a schedule, with costs capped and every result reproducible.
- Web, mobile and archiving
- Android, iOS, web and browser-extension software that does its work on the device, and preservation-grade web archives.
How we work: audit, consult, set up
No two organisations have the same data, rules or in-house skills, so we don't sell one box. We fit the work to what you need and what your team can operate.
Audit
We look at the data you hold, where it lives today, who can reach it, the regulations you work under and what your own team can operate. You get a written finding, whether or not you go further with us.
Consult
We recommend what fits your needs and your capabilities: your cloud account or your own hardware, which models, which work to start with. If an off-the-shelf tool already meets your rules, we say so.
Set up
A pilot in your environment on a bounded set of your documents, tested by your own people on real questions, then rollout with single sign-on and audit export. Your security team can inspect everything.
Support
Your team owns the installation. We stay on for updates, which you apply on your schedule, and for the next piece of work if there is one.
Four questions to ask any AI vendor, us included
Ask them before anything else. If the answers are vague, the product isn't ready for sensitive work, no matter how good the demo looks.
- Where is our data processed and stored?
- Our answer: In your cloud account, in the region you choose, or on your own hardware.
- Who holds the encryption keys?
- Our answer: You do, in your own key management service. We never see them.
- Is our data ever used to train a model?
- Our answer: No. Not by us, and not by the model provider.
- Can this run inside our own infrastructure?
- Our answer: Yes. A private deployment runs nowhere else.
For individuals and small teams
If you don't need your own installation, our products are ready to use today: Cognoir's hosted plans for asking questions of your documents, and Sitr Shield, a content filter that runs on your device. Same rules about data, no engagement needed.
See every productGot data you can't put in an online tool? Let's talk.
Tell us your organisation, where your data lives today and the regulations you work under. A founder replies, and the first conversation is with someone who can answer technical questions.