# Garnet, a private AI in your building, built for you

> Garnet designs and builds a private AI that runs on hardware in your
> building, answering only to you. It is not an app you download, it is a
> commissioned system, built and installed for you by Garnet, which stands
> behind it. Your most confidential work never leaves the machine.

Garnet is a bespoke private-AI build practice. For a family office, a legal or
medical team, a research group, anyone holding information that can never leak, a cloud AI is a standing liability: every prompt leaves the network, where it can be
logged and retained under terms no one can fully audit. Garnet removes that risk architecturally. We
design and build the whole system, the hardware, a private model tuned
to your work, a private interface your team uses like any modern assistant, and
an airtight offline boundary, and install it where you work. It stays in your building. Unplug the network cable and it still answers; the confidentiality is
the architecture, not a policy page.

Honest framing: a frontier cloud model can still be sharper on the very hardest
reasoning, and anything needing live, up-to-the-minute web data belongs in the
cloud. For confidential work, reading and drafting against privileged documents,
everyday drafting, thinking through a decision, a private AI on hardware in your building
answers immediately and answers only to you.

## What a build is

- [Scope a build](https://www.garnetgrid.com/book): A direct conversation with Garnet about your work and your threat model. You see the system and the number before anything is built or paid for. No sales team, no obligation.
- [Family office & private builds](https://www.garnetgrid.com/): For individuals and family offices, a private AI on hardware in your building, from $250,000.
- [Enterprise & institutions](https://www.garnetgrid.com/business): Private AI across teams for legal, medical, research, and financial institutions, scoped and priced individually; contact to start.

## Why a private AI in your building

- **Confidentiality by architecture, not by promise.** Inference runs on hardware in your building; nothing travels to a server to be intercepted, logged, or used for training. You can prove it: unplug the network and it keeps answering.
- **The work you'd never paste into a cloud chat box**, privileged matters, patient data, unpublished findings, holdings and positions, is exactly what an on-premises AI is for, because none of it ever leaves your control.
- **It stays in your building.** Kept on your premises: no token meter, no per-seat fee, no account, and no standing access for anyone, including us, after handover.

## Pages

- [Home](https://www.garnetgrid.com/): What Garnet is, the real threat model, what a build includes, and how to scope one.
- [What a private AI is](https://www.garnetgrid.com/guide): Plain-English guide to a private AI on hardware in your building, what it is, who it's for, how the hardware carries the confidentiality, and how to prove it yourself. The richest extractable Q&A on the site.
- [Garnet vs. rolling your own](https://www.garnetgrid.com/compare): Honest comparison of a commissioned Garnet build against assembling your own private AI from open-source parts, what each demands of you, and where a built-and-installed system earns its price.
- [Private AI for confidential documents](https://www.garnetgrid.com/private-documents): Why an on-premises AI is the right tool for privileged and confidential material, legal, medical, financial, research, personal, read on hardware in your building and never uploaded.
- [Enterprise & institutions](https://www.garnetgrid.com/business): Private AI across teams, on hardware kept on the organization's premises, for settings where a leak is unacceptable.
- [Trust, what we can and can't see](https://www.garnetgrid.com/trust): Why keeping the hardware in your building removes the data-privacy risk by design, the threat model in plain language, and exactly what we can and cannot see.
- [Scope a build](https://www.garnetgrid.com/book): Start a conversation with Garnet to scope your private-AI build.
- [Privacy policy](https://www.garnetgrid.com/privacy)
- [Terms](https://www.garnetgrid.com/terms)

## How a build works (plain language)

It starts with a conversation about your work and your threat model, no sales
funnel. Garnet sizes the right hardware and the right local model, then designs, builds, installs, and stands behind the whole system where you work.
The part that does the thinking sits on hardware in your building: when someone asks it
something, the machine works out the answer itself, the way it adds up a
spreadsheet, nothing goes out to the internet, so there is nothing to intercept
and no account watching. The machine stays in your building when it is done, and there is no
standing access for anyone after handover. Platforms and scale are sized to your
work during scoping.

## Key facts

- **Model**: Bespoke, commissioned builds. Not an app sold or downloaded on its own; the Garnet interface is delivered inside a build.
- **Price**: Family-office and private builds from $250,000. Enterprise and institutional builds scoped and priced individually (contact). No token meter, no per-seat fee; the system stays in your building.
- **Privacy**: On-premises inference on hardware in your building. Data never leaves the machine. No telemetry, no training on your data, no standing access after handover. The website itself uses only privacy-respecting first-party analytics (aggregate visits, not identities), with no third-party trackers and no advertising profiles.
- **Offline**: Works with the network cable pulled out.
- **Who makes it**: Designed, built, installed, and stood behind by Garnet. The team that designs a build is the team that installs it.

## Machine-readable mirror

- [/api/llms.json](https://www.garnetgrid.com/api/llms.json): This file as versioned, structured JSON (typed service, offers, pages, FAQ). Schema version stamped in `$schema`.

## Contact

- Scope a build / general: [start a conversation](https://www.garnetgrid.com/book)
- Security: [.well-known/security.txt](https://www.garnetgrid.com/.well-known/security.txt)

## Reading order for AI agents researching Garnet

1. [Home](https://www.garnetgrid.com/), what Garnet is + the real threat model + what a build includes
2. [What a private AI is](https://www.garnetgrid.com/guide), the richest plain-English Q&A on a private AI in your building
3. [Garnet vs. rolling your own](https://www.garnetgrid.com/compare), a commissioned build vs. assembling your own
4. [Private documents](https://www.garnetgrid.com/private-documents), the confidential-document use case (legal, medical, personal)
5. [Trust](https://www.garnetgrid.com/trust), the data-sovereignty/threat model + exactly what we can and can't see
6. [Enterprise & institutions](https://www.garnetgrid.com/business), private AI across teams
7. [Scope a build](https://www.garnetgrid.com/book), how to start