Knowledge Core
Knowledge Core
Knowledge Core is a searchable index of what your views are about. Indexing a view reads it, both the picture it draws and the definition behind it, and records its subject: the equipment on it, the area it covers, and the kind of display it is. Once a body of views is indexed, "show me the fractionator overview" is answered by searching that index instead of matching on view names, which is why the index is worth building across the views people actually go looking for rather than one at a time.
Beta
It is a separate capability, enabled per instance by IOTA through the KnowledgeCore feature flag. If it is not enabled for your instance, nothing on this page applies.
First run
Indexing is started one view at a time, from the Knowledge Core panel on the view toolbar, by someone who can already open that view. It can be stopped while it runs.
The panel shows live progress while a run is under way, and afterwards the last run's status, time and entity count. It lists the entities found in the view with their class, confidence and review status; selecting one highlights the components it appears in.
| Pass | What happens | What leaves your instance |
|---|---|---|
| Visual survey | The server opens the view as a browser would, captures one image of it, and asks a vision model what equipment, areas and view type it shows. | The image and the question asked about it. |
| Structured reading | The view's own definition (components, labels, tag paths, layout) is combined with the survey hints and read into a summary. | Labels, tag paths and layout. Not live values. |
What is kept
On your instance, and readable only by people who can already open the view:
- A summary of each view: purpose, area, type, level, status, confidence and review state.
- The entities found in it and how views link to one another.
- Each run's captured image and result.
Results the model is confident about are recorded as accepted and the rest as needing review, so a weak reading is visible as such.
What it does not do
It does not read anything outside the views you index, does not change your views, and does not widen what anyone can see: what the assistant surfaces from it is still filtered by each user's own permissions.
Related
- AI Guide: what the AI features do, and what is sent to the model
- AI Chat · AI Skills
- Feature Flags · Roles, Users, and Groups
Last reviewed 17 September 2026. Reviewed when the AI features change.