- Home
- Your data
Your data stays yours.
It runs in its own private setup in the US, encrypted and never used to train AI. Nothing goes out without a person's yes, and you own what we build.
Sending this to IT? Everything they'll ask is below, in 11 numbered sections.
Your data · Short version
For your IT person.
Have them call us, or send your security questionnaire. We'll answer every line. Want the paperwork first? This page as a PDF, the questionnaire answered and a BAA if you handle health records.
01
Where your data lives
- Single-tenant. Your brain runs in its own environment. It doesn't share a database, storage or index with any other client. We set up its accounts in your company's name, so you own them from day one.
- US-hosted. Data stays in US cloud regions. AI requests run in the US only.
- Yours to take. If we part ways, we hand over all access, or delete your data. See section 07.
02
The AI models: never used for training
Not ours, not anyone's. We use Anthropic and other leading AI models, always under business terms that don't train on your data. It's written into our agreement with you.
What each provider says, with links
Anthropic, maker of Claude:
"By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API...) to train our models."
Anthropic Privacy Center (opens in a new tab). The only exception they list is feedback you send or opt into. We never do either.
OpenAI, maker of ChatGPT:
"By default, we do not use data from ... the API platform, including inputs or outputs, for training or improving models."
OpenAI business data (opens in a new tab). We use OpenAI only through the API or business accounts, never personal ones.
Amazon Bedrock:
Model providers "don't have access to Amazon Bedrock logs or to customer prompts and completions."
AWS Bedrock data protection (opens in a new tab). Applies when we run the model in your AWS account.
Retention. Some AI providers keep requests for a short time to check for abuse, up to 30 days on some newer models (AWS (opens in a new tab)). Where your rules require it, we choose models and settings with zero retention.
03
Encryption
- In transit: TLS 1.2 or higher on every connection. AWS requires TLS 1.2 and recommends 1.3 for its services.Source: AWS (opens in a new tab)
- At rest: databases, files and backups are encrypted with AES-256. Our database runs on Supabase, which encrypts all data at rest with AES-256 and uses TLS 1.2 or higher in transit.Source: Supabase security (opens in a new tab)
04
What we connect
- You choose. You pick the systems, inboxes and folders we connect. Leave out anything you want, and turn any connection off later.
- Read-only to start. We ask for read access first. Write access, like drafting into your accounting system, comes later, one job at a time, with your OK.
- Your credentials, your control. Connections use the official sign-in and permission screens your software provides, wherever it offers them. You can revoke them in that software at any time.
05
Who sees what
- Your team. The brain answers by role. A tech can ask about a job and never see the books. A bookkeeper can see invoices and not payroll. You set the rules, and you can change them.
- Second Shift. Access is by role. Only Mac, plus our machine learning engineer on a complex build, gets in, and only to what setup or a fix needs.
- Nobody else. We don't sell or share your data.
06
A person approves
The brain answers your team's questions. It drafts follow-ups, reminders, reports and replies. Nothing goes to a customer, client, carrier or agency until a person on your team approves it. Every answer shows where it came from. When the brain isn't sure, it flags a person instead of guessing.
Sample · Draft, not sent
Hi Ms. Park, we have your W-2s and the 1099 from your bank. Could you send the brokerage statement by Friday?
AI drafts. Your people decide.
07
You own it
- Yours. You own your data, the brain and what we build for you. The accounts it runs in are in your company's name.
- No black box. We show you how every piece works.
- If we part ways, you choose. We hand over all access: the accounts, the setup, the code, the connections and the documents. Or we delete all your data and confirm it in writing.
08
If something goes wrong
We'll tell you within 8 hours of finding any security incident that touches your data, with what happened and what we're doing about it.
09
Companies we rely on
Every company that touches your data on our side, and what we use it for.
| Company | What for |
|---|---|
| Railway | Hosting |
| Supabase | Database and storage |
| Amazon Web Services, including Amazon Bedrock | Cloud services and AI models |
| Anthropic | AI models (Claude) |
| OpenAI | AI models |
| Google Workspace | Email and documents |
| GitHub | Code |
| Sentry | Error monitoring |
10
Certifications
- SOC 2. We don't hold SOC 2 yet. The companies we build on do, including AWS, Supabase and Anthropic.
- Your questionnaire. Send it over. We'll fill it out, line by line.
- Health records. We sign a business associate agreement (BAA).
11
Firms with rules of their own
We don't give legal, tax or compliance advice, and your duty stays yours. Here's what we give you to meet it.
- Law firm: ABA Formal Opinion 512 on AI and client confidentiality. What we give you: never used for training, single-tenant, sources on every answer, and a person approves everything that goes out.
- Accounting or tax firm: IRC 7216 and the FTC Safeguards Rule. What we give you: read-only to start, access by role, and a vendor you can oversee, with this page and our questionnaire answers.
- Insurance agency: only licensed staff advise or bind. What we give you: the brain drafts. Licensed staff decide.
- Financial advisor: SEC books-and-records rules. What we give you: drafts go out through your own archived systems, so your archive keeps them.
- Anyone with health records, such as PI firms: HIPAA. What we give you: we sign a business associate agreement (BAA).
