Briar

Privacy

Your clients' files do not become somebody's training data.

What Briar does with the information in your mailbox and your documents, what leaves the machine it runs on, what does not, and who else is involved. Written so you can check it rather than trust it.

Updated 7 min read

The short answer

Client documents are read on hardware dedicated to your practice and are not sent to an AI provider. Nothing Briar handles is used to train any model, and your book is never shared with another agent or sold. The parts that do use a hosted model are listed below by name rather than glossed over.

Where your client documents are read

Carrier PDFs, scans, photographed pages and the overnight pass that reads client files all run on a model installed on the machine Briar runs on for your practice. That machine is dedicated to you — it is not a shared, multi-tenant AI service, and your documents are not queued alongside anybody else's.

There is deliberately no hosted fallback on that path. If the local model is busy or unavailable, the work waits — it does not quietly get sent somewhere else instead. That decision matters more than it sounds: a fallback fires at the worst possible moment, unattended, and it would send exactly the information the arrangement exists to protect.

The specific hardware and where it is operated are set out in your agreement and confirmed during onboarding, so you know what you are relying on before any client information reaches it.

What is never done with your information

What does still leave, and where it goes

Briar routes each job to whichever model measures best at it, and two of those still run on hosted providers. Publishing the boundary is the point — you should be able to see where it falls rather than take our word for it. The specific providers, and the terms we hold them to, are named in your agreement and available on request.

What Briar is doingWhere it runsWhat is involved
Reading a carrier document, a scan or a photoYour dedicated hardwareThe whole page. It does not leave.
The overnight pass that reads client files for opportunitiesYour dedicated hardwareClient details and policy values. They do not leave.
Working out what you meant when you type, and answering questions about your bookA hosted AI provider, under contractYour own messages, and the records needed to answer them — which can name a client.
Deciding whether an incoming email needs youA hosted AI provider, under contractSender name, sender address and subject line. Not the body.
Writing a client letter or building a proposal documentA hosted AI provider, under contractThe text being written, and the facts it is grounded in.
Your mailbox connectionGoogle, via OAuth you grant and can revokeBriar holds a per-agent credential, encrypted at rest. You can withdraw access from your Google account at any time.

How the information is protected inside Briar

  1. 01

    One agent cannot see another's book, and it is not the application deciding that

    Separation is enforced by PostgreSQL row-level security. Every tenant table carries a policy the database applies to every query — 52 tables, all of them forced, verified automatically rather than by review. A bug in the application code cannot leak across agents, because the application is not what is enforcing it.

  2. 02

    Documents are encrypted at rest and stored once

    Each file is stored a single time, keyed by its hash, so the same document arriving twice does not become two copies to secure.

  3. 03

    Anything that arrives is treated as data, never as an instruction

    Email bodies, attachments and web pages cannot tell Briar to do anything. Document reading returns typed fields only — there is no free-text field in the output — so an instruction hidden inside a PDF has nowhere to live even if a model reads it.

  4. 04

    Nothing leaves without you releasing it

    An inbound email can never cause an outbound one. No lane can send a document or an email on the first message: you are shown the exact wording and the exact recipient, and nothing moves until you reply.

  5. 05

    Every automated decision is logged with its evidence

    Where Briar filed something on its own judgement, it records what it filed and the data point that justified it, so "why did that end up there" is answerable months later rather than being a matter of opinion.

  6. 06

    Some information is refused outright over chat

    Social security numbers, bank details, medical information and applications are not sent over a chat channel — to anyone, including you. That is a rule about the channel, not about who is asking.

What Briar stores about you, as the customer

Keeping it, and getting it back

Questions, and the honest caveats

If you want something on this page evidenced before you connect a mailbox, ask — the specifics above are stated precisely so they can be checked rather than admired. Write to hello@meetbriar.com.

This page is a description of how Briar works today and will be updated when that changes; the date at the top is when it was last checked against the running system. Briar prepares work for licensed insurance professionals and does not provide insurance advice.

Questions

Asked and answered.

Is my client data used to train AI models?
No. Nothing Briar handles is used to train a model, by Briar or by the providers it uses. The hosted accounts Briar runs on have training and prompt retention switched off in the provider's own settings, and those settings are re-checked whenever a provider changes.
Do my clients' documents get sent to an AI company?
No. Carrier documents, scans and photographed pages are read by a model running on hardware dedicated to your practice, and there is no fallback that sends them out if that model is unavailable. The overnight pass that reads client files works the same way.
What does still get sent to a third party?
Two jobs. Working out what you meant and answering questions about your book runs on a hosted model, carrying your own messages and the records needed to answer. Writing a client letter runs on a hosted model too. Both are named on this page, with the provider, rather than left vague.
Can another agent using Briar see my book?
No. Separation is enforced by PostgreSQL row-level security across 52 tables, all of them forced, so the database rejects a cross-agent read regardless of what the application asks for. Coverage is verified automatically rather than by review.
Can I get my data out, or delete it?
Yes to both. You can request an export of your book in a loadable form at any time, and deletion on termination, confirmed in writing. You can also disconnect the mailbox directly from your own Google account without involving us.
Is Briar HIPAA or SOC 2 compliant?
Briar does not hold a SOC 2 report or an ISO 27001 certificate, and does not claim to. A business associate agreement with one hosted provider has been requested and is not signed yet; this page will say so when it is. The controls Briar does have are described above in checkable terms.

Early access

Briar does the admin. You do the part only a licensed human can.

Briar is onboarding a small number of independent life agents. If you are spending your week on administration instead of clients, get in touch and tell us what your week actually looks like.