Released 14 September 2026

What changed in 0.59.7

A model run becomes a record you can trust: a finished run is read or copied, never edited, and one APPLY creates everything it produced — saying plainly what it left out. The AI assistants know your database, Integrate ▸ Mappings filters by run, and a slow merge step that could hold Extract for minutes is gone.

84commits since production
7new capabilities
8defects fixed

Building a model

New

A finished run is history — view it, clone it, or continue an unfinished one

VIEW RUN walks a finished run with every option locked; the files still open to read. CLONE starts a new run with its own name and identifiers, holding a copy of the model and the source files, so the two can be compared side by side. CONTINUE picks up a run you left halfway.

Model ▸ Generate ▸ RUNS
Gone

EDIT on a finished run

Changing part of a finished run left it half old and half new, with nothing saying which. Clone it instead — the original stays exactly as it was.

Model ▸ Generate ▸ RUNS
New

One APPLY creates everything, and says what it left out

The model files, the test data and the competence questions are created together behind a single confirmation. Anything that could not be created is listed with the reason, and a run is applied once.

Model ▸ Generate ▸ Result
New

Only questions that answer are saved

A competence question is saved to your library only if it returned values against the run's test data. The ones that did not are named on the Result step, and marked on Verify.

Model ▸ Generate ▸ Verify, Result
New

Delete a source file from a run

An uploaded file can be removed from a run you are working on, after a confirmation that names it. What the wizard generates is read-only there — edit it where it lives, in Model or Integrate.

Model ▸ Generate ▸ Sources
Fixed

Extract no longer sits for minutes bringing findings together

The concepts from each part of your documents are merged directly, and the AI is asked only which names mean the same thing. A reply that short cannot run out of room, so the step no longer retries again and again.

Model ▸ Generate ▸ Extract
Fixed

Deleting a run removes everything it put in your database

Including the classes, shapes and files of a run that came with a template, which used to stay behind and appear as duplicates in Explore.

Model ▸ Generate ▸ RUNS ▸ DELETE

AI assistants

New

Assistants that know your database

Every assistant is given your database's metadata, keeps its conversations to one user in one database, and the Query assistant sees the query on your screen.

Every screen ▸ AI Assistant

Integrate

New

Filter mappings by the run that made them

The same filters as Ontologies and Shapes: your own mappings, or those from a model run, one chip per run. The buttons carry words, like the rest of the app.

Integrate ▸ Mappings
Fixed

A new mapping keeps only the source you picked

Switching the source on a new mapping sent the first one along too, so saving refused fields that were not in the dialog.

Integrate ▸ Mappings ▸ CREATE MAPPING

Explore

Fixed

One “Default Graph”

It appeared both as a quick filter and as a graph to tick; ticking the second made Traverse fail. It is now a quick filter only.

Explore ▸ Graphs
Fixed

“Ontologies Only” says why it is unavailable

On a database with no reference ontologies it did nothing; it now says it would show the same as All.

Explore ▸ Traverse ▸ Ontologies

Engine and administration

New

Faster materialization on large databases

Inferred facts are written through the same bulk path the importer uses.

Connect ▸ database card ⋮ ▸ Config ▸ Reasoner
Fixed

Clear Inferred no longer freezes the app

It removes inferred facts in bounded chunks instead of one closure-sized batch.

Connect ▸ database card ⋮ ▸ Config ▸ Reasoner
Fixed

One unreadable database no longer blanks the list

The list shows every database it can read, and Admin ▸ Databases and Users open without walking every store file.

Connect, Admin ▸ Databases
Fixed

A taxonomy node that fails to save says so

The error is shown in the dialog and the tree stays as it was.

Model ▸ Taxonomies