Trustees & Fiduciaries
Source-grounded AI for confidential fiduciary records.
Trust deeds, registers, minutes, correspondence, and compliance files — retrievable in seconds, cited to the page, separated by mandate, and never outside your custody.
The working reality
Fiduciary work is document-heavy and precision-critical. Senior time disappears into finding the controlling clause, the historical resolution, the letter that explains why a structure looks the way it does. A hallucinated answer is worse than no answer — and client confidentiality rules out casual use of public AI tools. The system is built for the opposite: retrieve, cite, and refuse when the record is silent.
Where senior time comes back
The controlling clause
The governing language across a deed, its amendments, and later resolutions — located and cited, instead of re-read end to end.
Mandate history
The documentary history of a mandate reconstructed before a meeting or an audit: what was decided, by whom, and on what paper.
KYC / AML research
Structured search across your own client and compliance records — supporting the file, never replacing the officer’s assessment.
Drafting from sources
Correspondence and memoranda drafted from cited material, ready for professional review — a first draft with its evidence attached.
What the system does
- Retrieves the governing language across deeds, amendments, and resolutions — with citations
- Reconstructs the documentary history of a mandate before a meeting or an audit
- Drafts correspondence and memoranda from sourced material, ready for professional review
- Supports KYC/AML file research across your own records
- Enforces mandate-level separation: staff see only the matters they are engaged on
- Refuses to answer where your records give no basis — visibly, in the log
Matter separation and evidence
Confidentiality between mandates is a control, not a convention. Staff see only the matters they are engaged on; cross-mandate retrieval simply does not resolve. Every retrieval and every answer is logged per mandate, with the model version and the sources it used — the evidence you can put in front of a regulator, an auditor, or a beneficiary asking how an answer was reached. The system is tested for exactly this before you rely on it: groundedness, refusal on unsupported claims, prompt-injection resistance, and access-control separation.
How an engagement runs
- Confidential architecture review1–2 weeks, fixed scope We map the mandate structure, the sensitivity classes, and the right custody model — on paper, before any records are ingested.
- Proof node2–6 weeks A sanitised set of matters runs on real infrastructure. You see citations, refusals, and per-mandate audit logs on your own material — then decide to stop, expand, or productionise.
- Production system and managed operation6–16 weeks, then ongoing Warrantied hardware, hardened deployment, mandate-level roles, monitored backups, an evaluation pack, and a support contract that keeps retrieval dependable.
Custody posture
Typically on-premises with strict role separation; air-gapped where the client base warrants it. Every retrieval and every answer is logged per mandate — evidence you can show a regulator or an auditing beneficiary.
What we deliberately do not do
The system retrieves and drafts with sources. It does not replace fiduciary or legal judgement, it does not decide anything on its own, and it refuses where your records give it no basis. Your professionals remain the decision-makers on every matter; the system makes the paper behind their judgement faster to reach and easier to cite.