No training
Nothing in a Keep is used to train any model. Client financial data should never be an AI vendor's training set.
Home / Zero-data-retention AI
DefinitionModel providers do not keep client data for training; the firm controls retention and deletion; nothing is shared across trusts; only a fingerprint leaves the firm boundary.
Zero-data-retention AI for fiduciaries means three things at once: model providers do not keep client data for training; the firm controls retention and deletion; nothing is shared across trusts. Only a fingerprint leaves the firm boundary. It does not mean r.Team keeps no fiduciary records.
Nothing in a Keep is used to train any model. Client financial data should never be an AI vendor's training set.
The trust company sets retention and deletion. Closing a trust, or a family deleting a document, removes it everywhere in one action.
A tamper-evident record the firm is obliged to keep. The proposal and the human decision stay distinct.
Families are already asking public AI assistants about statements, the trust, and their own documents. For a trust company, that is a boundary problem: account numbers, positions, beneficiary names, and the governing instrument leave under consumer terms the firm never reviewed and cannot enforce.
Data leaves the boundary under consumer terms the trust company never reviewed. There is no firm record of what was asked or answered. The trustee is cut out.
Keep runs inside a data boundary the trust company controls. Nothing a family places in a Keep is used to train any model, shared with any other trust, or readable by anyone the family has not authorized. The trust company's own agreements govern the data.
These are the controls already published on Keep and the Platform. Architecture, isolation, and subprocessors are reviewed in diligence under NDA. We do not publish a certification badge set on this site.
Nothing in a Keep is used to train any model. The trust company's own agreements govern the data, not a consumer assistant's terms of service.
Each Family Trust Account has its own sealed space. Separation is enforced at the data layer, not by a rule the model is asked to follow.
The trust company sets retention. Closing a trust, or a family deleting a document, removes it everywhere in one action.
Account numbers are held as last-four only. Identifiers the model does not need are never stored.
Client data stays inside the firm boundary. The public record carries a fingerprint, not a file. Only a 32-byte fingerprint is published.
If you leave, your data leaves with you. Keep attaches to what you run today. It does not replace trust accounting, custody, or the portfolio system.
Zero-data-retention here is about model providers and training pipelines — not about erasing the fiduciary file. Keep writes every exchange to the trust's own record. The Record is a tamper-evident record — a hash-chained log — so an examiner can check later. That is the record the firm is obliged to keep.
Every question, the sources used, and the answer are written to the trust's own record, alongside the officer's decisions. The proposal and the decision are distinct, and both are kept.
Every fiduciary event is written to a per-account hash chain. The human decision is part of the record. The engine's proposal is part of the record. They are distinct.
Model calls are recorded with the agent, the account scope, the policy version, and the prompt context hash. Inference traffic becomes an auditable ledger, not a black box.
The same boundary runs through Keep, the Platform, and the Record. The copilot still prepares, cites, and stops. The named officer decides.
Keep runs inside a data boundary the trust company controls. One Keep per Family Trust Account. Explains and cites. Never advises. Private by design: instruments stay inside the firm boundary.
Keep →Client data stays inside the firm boundary. No client names, account numbers, positions, instruments, or rationales leave it. Architecture, isolation, and subprocessors are reviewed in diligence under NDA.
Platform →A tamper-evident record. Hash-chained events. A fingerprint — never a client file. Change an earlier entry and the break is visible.
The Record on the Platform →Client financial data should never be an AI vendor's training set. Model choice and governance without committing data to any single AI lab. Documents held close, zero egress, on the same platform.
Why Cloudflare →Prepares, explains, and proposes beside the team. It does not decide.
One governed doorway to the systems you already run. Read first. Write only after human approval.
A tamper-evident record. Hash-chained events. A fingerprint — never a client file.
A fiduciary copilot still reads, prepares, and cites. Nothing happens until a person approves. See Keep, the Platform, and Why Cloudflare for the stack.
Related: Keep · Platform · What is a fiduciary copilot? · AI for bank trust departments and trust companies · What is a Trust Administrator Pod? · Why Cloudflare · Privacy Policy · Vocabulary
Short answers from the published site. Open a question to read it.
No. Keep writes every exchange to the trust's own record. The Record is a tamper-evident record — a hash-chained log of every decision — that an examiner can check later. The proposal and the human decision are distinct, and both are kept. That record is what the firm is obliged to maintain.
Nothing a family places in a Keep is used to train any model, shared with any other trust, or readable by anyone the family has not authorized. The trust company's own agreements govern the data, not a consumer assistant's terms of service. Client financial data should never be an AI vendor's training set.
Client data stays inside the firm boundary. The public record carries a fingerprint, not a file. No client names, account numbers, positions, instruments, or rationales leave the firm boundary. Only a 32-byte fingerprint is published.
The trust company sets retention. Closing a trust, or a family deleting a document, removes it everywhere in one action. Account numbers are held as last-four only; identifiers the model does not need are never stored. If you leave, your data leaves with you.
One boundary per trust. Each Family Trust Account has its own sealed space. The separation is enforced at the data layer, not by a rule the model is asked to follow. Instruments, accounts, and rationales stay inside the firm's boundary. Nothing about one client's trust informs another firm's work.
No. The website Privacy Policy explains how Got-Data Inc. collects, uses, and protects information when you visit r.team or submit forms. This page is the product posture: model providers do not keep client data for training; the firm controls retention and deletion; the Record is kept on purpose.
A focused demo of the Reasoning Engine, the Bridge, and the Record — or start a conversation.
A named officer decides, and every decision is recorded.