Trust and transparency

AI drafts. The archivist decides.

Nothing reaches a catalogue, an export or the public without a named person accepting it.

Start free Talk to us

EU-resident AI processing · Human review on every field

The AI reasoning field on an Archivers.ai record for match footage, explaining that the title and date came from on-screen graphics and the people were identified from the commentary, beside fields for detected personal data, access rights and sensitivity level
Evidence

Read why, not just what.

Each drafted field carries a written citation: what the AI saw on the page, or a plain statement that it inferred the value with no direct evidence.

  • Evidence cited field by field inside the record
  • Reasoning you can open and challenge
  • Personal data and sensitivity raised early
An unreviewed Archivers.ai record for a printed index page of Coroners’ Inquests 1732–1836, with the drafted title, description and subject tags shown directly above the scanned source page they were read from, and an Accept button waiting
Retroconversion

Checked against the original page.

Converting a legacy catalogue raises a harder question than accuracy: was anything on the page missed?

  • Record and scanned page, side by side
  • Pages are signed off, not only fields
  • The source link stays on the record
Digitisation projects
Verification

Four controls on a legacy conversion.

Built for how volunteer-heavy services actually staff this work.

Side-by-side check

Every drafted record can be read against the scanned page it came from.

Page coverage

A reviewer confirms nothing on a page was missed, not only that fields look right.

Two-person review

A volunteer checks against the source; an archivist gives the final sign-off.

Permanent citation

Each record keeps a link back to its exact source page, for good.

Provenance

What the AI did, written down.

The record of the decision travels with the record itself, into whatever holds it next.

PREMIS events

in BagIt exports

Each AI action and each human review is written as a preservation event with a timestamp, an agent and an outcome.

EAD3 maintenance history

on finding aids

Finding aids record that AI-assisted cataloguing was used, when it ran, and which software agent produced the draft.

Audit trail

in the app

Processing events, edits, overrides and approvals are logged, attributed and exportable — so you can show who did what, and when.

EU data centres for AI processing

No-training commitments from our AI vendors

Your notes shape the draft, not the reverse

Trust page sub-processors and exit

Judgement stays with the archivist.

Start free with 30 credits a month, or talk to us about an institutional pilot.

Start free Talk to us