Model Register¶
The Model Register is the inventory of every AI system you govern: models, LLM apps and agent systems. It's where an auditor's first question, what AI do you run, and who's accountable for it?, gets its answer. Each row opens the system's dossier.
Press g then r to get here. Register a system adds one from the dashboard; to add one from
code, see Getting Started.
The register table¶
Which columns lead depends on your lens. Available columns:
| Column | Shows |
|---|---|
| Model, Type, Status, Owner, Registered | Identity and lifecycle |
| Trust score | The system's current Trust Score |
| EU AI Act tier | The risk classification recorded for the system |
| Lifecycle stage | Where the system is in its lifecycle |
| Open findings | Unresolved findings against the system |
| Evidence | Whether approved evidence exists |
| Last attested | When the system was last attested |
| Drift, Fairness gap | Latest drift status and fairness gap |
| Last seen | When data last arrived from the system |
Filters¶
Search by name, plus Risk tier, Lifecycle, Drift, Findings, Evidence (Has approved / Missing evidence), Status (Active, Deprecated, Archived), Owner and Tag. Filters are kept in the page address, so you can share a filtered view by sharing the link. Clear filters resets them.
Toolbar¶
- Saved views: save the current filters and columns (how)
- Columns: choose which columns to show, or Reset to default
- Export CSV: exports every row that matches your filters, not just the visible page
- Compare systems, or select rows and Compare N selected: compare systems side by side
The dossier¶
Click a system to open its dossier in a slide-over. The page address includes the system, so you can link to a dossier directly.
Governance verdict¶
Four verdict cards summarise the system's state in plain language. Each has How was this computed? and, for analysts and administrators, Mark reviewed. Marking a verdict reviewed records who reviewed it and when. It's refused if the underlying figures have changed since the page loaded, so nobody signs off on numbers they didn't see.
Assurance timeline¶
Assurance events for the system (audits, drift reports, reviews and decisions) in date order.
Signals¶
| Tab | Contents |
|---|---|
| Drift | Latest drift results for the system |
| Fairness | Latest bias audit, including "Recommended mitigations, as the audit recorded them" |
| Explainability | Recent explanations for the system's predictions |
| LLM & agents | LLM usage, safety and agent activity for LLM-based systems |
| Lineage & versions | Every registered version, with Compare versions and see the code diff |
| Model card | The system's model card: card version, publication status and freshness |
A model card marked Stale — needs refresh says why under Why this model card is marked stale: a newer source commit was registered, or a newer fairness audit or drift report was recorded, after the card was last updated.
Refresh Model Card updates the card from what WhiteBoxXAI already holds: the model's current
version and baseline metrics, its latest drift report, and its latest bias audit results. If the
system has no card yet, it creates one. Nothing is re-typed by hand, and the card's freshness updates
once the refresh completes. From the API: POST /api/v1/compliance/model-cards/refresh/{model_id}.
Findings, coverage and history¶
- Open findings for the system
- Control coverage: N of M in-scope controls with evidence (a count, never a percentage)
- Trust score history, with its three components
- Audit trail: every recorded action on this system
System settings¶
- Edit details: description, status and tags, then Save changes
- Archive / Restore: archiving asks "Archive this model?". Archived systems keep their history, stop counting towards your plan's Governed AI Systems, and can be restored later.
Versions and code diffs¶
Register a new version of a system by registering it under the same name with a new version string. It stays one Governed AI System. From Lineage & versions, compare two versions to see what changed between them, including the code changes when the versions were registered with Git context. Model-diff architecture reviews are included on Business Cloud and Enterprise Edition. See GitHub Diff Reviews.
From code¶
client.models.list(search="credit", limit=50)
client.models.get(model_id)
client.models.get_versions("credit_scoring")
client.models.list_git_versions(model_id) # Git-backed versions
client.models.compare_versions(version1_id, version2_id) # ids from list_git_versions
client.models.archive(model_id)
client.models.restore(model_id)
See SDK & API.