AI Intel
What an organisation administrator can set for AI in Virtik, where each setting takes effect, and which AI features it does and does not govern.
- Version
- 0.2
- Zuletzt aktualisiert
Dokumentation durchsuchen
Auf dieser Seite
Dieses Dokument wird derzeit nur auf Englisch veröffentlicht. Englische Originalfassung lesen
AI Intel is Virtik's name for governed AI execution: AI that runs under explicit operating modes, organisation policy and budget control. It has no screen of its own in the application's navigation. It is the set of policy settings that the AI gateway applies to the AI features that use it.
What it governs
The gateway applies this policy to ChatVRTK, to the Command Dock's assistant, and to the AI helpers in Prevention and Training. It does not govern camera analysis in Detection, which calls a vision provider directly under its own checks, nor the scripted assistant screens described in AI governance.
What an administrator can set
| Setting | What it controls | Where it applies |
|---|---|---|
| Providers | Which AI providers the organisation connects. Members can also add a personal key where the organisation allows it, and a plan can include Virtik's own managed allowance. | Per organisation |
| Default operating mode | The mode used when a request asks for Auto, and whether people may choose a lower one. | Per organisation |
| Budget and rate limit | A spending limit for the organisation and, separately, per user, over a daily or monthly period or none, as soft or hard; and a limit on requests a minute. | Per provider |
| Allowed data classes | Which classes of data may be sent to a provider. | Per provider |
| Approvals | Not a setting: an action in the critical risk class waits for an organisation administrator to approve or reject it. | Per organisation |
Only an organisation administrator can change this policy.
Things to know
- A soft budget warns in the API, not on screen. A soft budget lets the request through and returns a warning that the usage panel shows only as a coloured bar. A hard budget refuses.
- Cost figures are estimates, calculated from list prices.
- The class of data is declared by the feature that sends it, not detected from the content. ChatVRTK and Training declare "internal", and Prevention declares "confidential".
- The organisation-wide data policy has no screen. By default, confidential data is set to "require approval", which is not implemented (the request is not sent), and restricted data is denied.
- A minimum mode per module exists in the database only. The settings screen offers a default mode and the choice to let people lower it.
- Virtik's managed allowance is metered. When it is used up, requests are refused with a message that says so.