How your data is protected, and how AI is used

Only a person Prairie Forge or one of your company's Leads has added can sign in. When you sign in, a one-time button and code are emailed to you (a counter gets theirs by text or email, in the OnHand app); either signs you in, and both expire in an hour. Blueprint never gives you a password or asks for one, so there is none to remember or have phished.

After 8 hours without activity you are signed out and asked for a new code.

Each person has a role — Lead, Member, Viewer or Counter — and a Lead can limit what a person sees to the tools, companies or locations they work with. A paused person keeps their setup but cannot open anything until a Lead resumes them.

Each company's data lives in its own schema in the database, and every read is scoped to the company you are signed in to. Nothing crosses between companies, structurally.

When someone on the Prairie Forge team reaches your data through the Claude connector, Blueprint attempts to record the query with who ran it and the reason they gave. When a staff member requests a download link for a restricted original document, Blueprint attempts to record that request with who asked, and their reason when one is given. A failed record does not stop the request. A staff member opening a screen or previewing a document inside Blueprint is not recorded person by person today.

Application data is stored in a Supabase-hosted PostgreSQL database with encryption at rest and in transit. Supabase runs on AWS infrastructure. Uploaded files — statements, transcripts, filed documents — are stored in Supabase Storage, also encrypted.

The web application is hosted on Vercel, which serves the application code; it does not hold your data.

Social Security, account and card numbers are masked before a document is stored in the notebook, and the reader leaves salaries, terminations, health details and non-public deals out of the record it keeps. A document the checks cannot clear is held for a person to look at, never slipped in.

The source code is kept in a private repository accessible only to Prairie Forge. It contains no client data and no credentials; those are managed as secured settings on the hosting platform.

Blueprint uses Anthropic's Claude through its commercial API. Ada reads your company's notebook and books to answer a question, and the readers that take in documents use it to write a document's record.

The AI is instructed to use only what is in your Blueprint — to flag a gap rather than fill it. Sensitive specifics are left out of what it stores.

No. Anthropic's commercial terms govern the API Blueprint uses: inputs and outputs are not used to train its models, and traffic is kept only briefly for abuse and safety monitoring. This is different from the consumer Claude apps.

A number Blueprint shows ties to a loaded source or it is marked as not tying — on every surface that shows that month. An analysis Ada builds carries how it was built. Nothing AI-generated is delivered by Prairie Forge as its own work without the engagement team's review, and your team is asked to confirm what the final deliverables say about your business.

When the Blueprint connector is enabled, Claude can read your Blueprint through a small set of read-only tools, scoped to your company on every call, and save your own work into your own folder. The text of a search is logged as a fingerprint, not in plain words.