Released 11 September 2026

What changed in 0.59.6

Taxonomies now work end to end: create one, fill it, import several at once from any RDF format. A finished model run can be read without re-running it, Microsoft Azure OpenAI joins the AI providers, and our W3C SHACL conformance report is published in full — 98 cases out of 98.

7defects fixed
4new capabilities
98/98W3C SHACL Core cases

Taxonomies

Two faults that hid each other. The second was unreachable until the first was fixed, so nobody had ever been able to hit it.

Fixed

Creating a taxonomy produces one you can open

A new taxonomy was written to disk and then left out of the list, so the screen looked unchanged and the taxonomy was unreachable for good. Any taxonomy you created before and could not find is back in your list.

Model ▸ Taxonomies ▸ CREATE TAXONOMY
Fixed

Adding, renaming and deleting nodes

Every one of these answered “not found” on a taxonomy you had made yourself, because the list handed the screen an identifier the save could not accept.

Model ▸ Taxonomies
New

Import taxonomies

Several SKOS files at once, in any RDF serialisation, listed with format and size before you commit. A file holding no concept scheme is refused rather than saved somewhere nothing lists it.

Model ▸ Taxonomies ▸ IMPORT

Queries and APIs

Fixed

TRY IT NOW on a saved API

It failed every time it was pressed, as did running a saved query from the library. A viewer is still correctly refused an update.

Query ▸ API
Fixed

Clearer limits when a query is stopped

A query stopped by the memory budget states the figure in megabytes rather than rounding it to “0.0 GB”, and one stopped by the time limit is no longer reported as a server fault.

Query ▸ Editor

Building a model

New

Read a finished run without re-running it

VIEW RUN walks the steps of a completed run and executes nothing; NEXT moves you along rather than starting work. EDIT opens the same run to change it.

Model ▸ Generate ▸ RUNS
Fixed

A working run no longer looks stuck

Extraction reads your files and then asks the model to bring the findings together. That second pass said nothing, so the screen held the last count and appeared frozen. It now says what it is doing, and a cancelled step no longer leaves its progress behind.

Model ▸ Generate ▸ WIZARD
Fixed

Findings appear once

What your sources disagreed on was shown twice; it now appears only on the step that produced it.

Model ▸ Generate ▸ Extract

AI providers

New

Microsoft Azure OpenAI

Use an Azure OpenAI resource with its endpoint, deployment name and API version. A connection that cannot reach the endpoint now says so plainly instead of showing a technical error.

Admin ▸ Settings ▸ AI

Standards, in the open

Claims about correctness are cheap. This one can be checked.

New

The W3C SHACL conformance report is published

Every approved case of the official W3C SHACL Core test suite, and our result on each one. RDF Studio passes 98 of 98. It is generated from a live run of the suite rather than written by hand.

Learn ▸ Guides ▸ W3C SHACL Conformance Report
Fixed

The manual describes validation correctly

It said validation was bounded by how much data you had. It runs inside the database engine, so the size of your database is no longer what decides whether a run finishes.

Documentation