Skip to main content

AI Organisation

Let the AI propose a fonds structure based on the accessions you've already ingested. Useful when:

  • You've accumulated dozens of unassigned accessions and need to organise them
  • You've inherited an unorganised archive and want a starting point
  • You want a second opinion on an arrangement you've drafted yourself

Available on Professional plans and above.

How it works​

The platform reads:

  • Every accession's title, description, creator field, date range
  • The categories and dominant subjects of items within each accession
  • Authority links extracted across the accession

…and proposes a fonds hierarchy that groups related accessions under common creators or themes.

The proposal is always a draft — nothing changes in your archive until you accept it.

Running an organise​

  1. Fonds in the sidebar.
  2. Click AI Organise (top right, near Manage).
  3. Pick which accessions to include in the analysis — by default, every unassigned accession. You can also include already-organised accessions for a re-do.
  4. Add organising hints (optional but recommended). Examples:
    • "Group by donor"
    • "Use ISAD(G) — sub-fonds for departments"
    • "These are all from the Smith family but cover three generations"
    • "Keep the existing top-level structure; only suggest new series within fonds"
  5. Click Generate proposal.

Processing takes 30–90 seconds depending on how many accessions are in scope.

Reviewing the proposal​

The proposal is shown as a diff view against your current structure:

  • Green nodes — new fonds, sub-fonds, series, or sub-series the AI proposes
  • Blue arrows — accessions the AI proposes to move into specific places
  • Grey — your existing structure, unchanged

You can:

  • Accept individual proposals — tick the ones you like
  • Edit titles and descriptions inline before accepting
  • Ignore the rest — anything you don't tick stays as-is

Click Apply selected to commit. Anything not selected stays Unassigned or in its current location.

What the AI is good at​

  • Identifying when several accessions clearly come from the same creator (matching surnames, organisation references, date overlaps)
  • Suggesting series groupings within a fond based on document type (correspondence vs. minutes vs. photographs)
  • Pulling out date ranges and using them as series boundaries

What it's less good at​

  • Provenance you haven't told it about. If your accession descriptions don't mention the donor, the model can't infer it.
  • Subtle institutional structures. A council's records might map to departments that no longer exist; without a hint, the AI will produce a thematic grouping instead.
  • Conflicting evidence. When two accessions look like they could belong to the same creator or to different ones, the model leans cautious — it'll propose them as sub-fonds rather than merge.

For these, use the hints field. The more specific you are, the better the proposal.

When to re-run​

  • After a big ingest — bring new accessions into the structure
  • After improving descriptions — better descriptions feed better proposals
  • When you've changed your mind about a top-level grouping — give a hint like "split the Council fond into two by decade"

Re-runs are idempotent: nothing is changed until you accept proposals from the new run.