AI settings

Admin, AI settings. Where a school decides what the assistant and the background AI features are allowed to do, and where the money side of that is capped.

The school switch

ai.enabled stops every model call for the school, on-demand and background, the moment it is turned off. Left unset, on-demand features work wherever an AI connector exists, and background AI runs only for schools with managed AI.

With no AI connector configured at all, the desk behaves exactly as it did before 0.19: no AI switch to find, nothing running in the background.

A mode per feature

Each AI feature — ticket triage, the reply draft, same-issue detection, the repair likely-outcome card, device fixes and the rest — has its own mode:

Mode	What it does
Off	The feature runs with no model, the same as with no AI connector
Shadow	The model runs and its output is recorded, but nobody sees it
Suggest	Shown as a ghost value or a card, with one-click accept, dismiss or undo
Auto	Applied without a person confirming it first

Auto unlocks only after a proven accuracy record for that feature at that school, and never for identity work — joiners, movers and leavers runbooks always stop at a person’s approval, however long a feature has been running cleanly elsewhere.

Pseudonymisation

On by default: names on the school’s roster, email addresses, phone numbers, IDs, device serials and street addresses are replaced with tokens before a call leaves and restored in the answer. It is always on for managed AI. For a school’s own connector, an administrator can switch it off here, which needs a step-up confirmation.

Cost controls

Managed AI’s fair-use cap is 8 calls per ticket per month, plus a school’s own daily budget, ai.dailyBudgetMicros (US$2 by default), set on this page. Usage against the plan’s monthly allowance shows on Admin, Licence.

Related

From 0.22.0, optional decision providers handle fixed classifications separately from the chat provider used for drafts. Decision requests always mask identifying data, and each model earns its own approval history. Laya has no trained Plugboard checkpoint in this release; Jev starts disabled. Review suggestions on your own data before considering automatic use.

• The assistant and AI features
• AI activity, the audited record of every call
• What we process, for pseudonymisation and where AI runs

Assistant command decisions (0.22.0)

Optional decision providers can recognise an existing assistant command before conversational reasoning. Exact commands keep their fast path. Permissions, record access and action reviews still apply; a classification never supplies an arbitrary tool or target. Each new model starts in shadow. Shown interpretations can be corrected without running another command.

Portal and catalogue choices (0.22.0)

Decision providers can match an existing help article or school service while someone describes a problem. Choosing a service fills its form; the person still completes and submits it. Existing selections are retained. Portal feedback is linked to the person’s own suggestion and fails safely if content changes.

Access and routine request matching have separate model records. Identity requests retain staff approvals and step-up. Model-based routine fulfilment needs both the school’s pre-approval and the answering model’s earned auto record; switching providers cannot inherit the old model’s accuracy.

School criteria and tags (0.22.0)

In Admin → Workflow → AI criteria, describe when each category, subcategory, priority, department, repair type or catalogue item fits. Use up to 200 characters and five example phrases of up to 120 characters each. Save and try it makes a metered sample decision without changing a ticket. The criteria follow your selected decision provider. Priorities still follow the impact/urgency matrix.

Admin → AI → School tags supports ten yes/no questions. Each starts in shadow. Suggestions offer Add label and Not relevant; applied tags are normal labels staff can remove. Every tag and actual model needs its own 50 human decisions at 90% accuracy over 60 days before auto can apply it. Changing the question or description starts a fresh record. Manually changing one label does not label all the other AI questions as false.

Local full-context triage with all ten tags is background work. On the isolated four-CPU staging VM, five warm synthetic runs measured 56.69s median / 57.13s p95; the bounded Laya deadline is 75s. Assistant intent and intake retain their 2.5s limits. These timings do not establish accuracy or production capacity.

Similar resolved tickets (0.22.0)

Triage can use up to ten resolved tickets that staff classified in the same school, campus and department. The review links each supporting record. If history disagrees with the decision model, staff must review the choice. History-only suggestions always require a person in this version. Support from history is shown separately from the model’s answer probability. Changed or reopened source tickets invalidate the review. An embedder is required; no training or decision provider is required for this history path.