AI models

One choice for every application

The AI features across the applications — chats, translation, analysis, content generation — work with a specific model. That model is chosen once, in your account, and applies to every application at the same time: there is nothing left to configure inside each one.

Who decides what

• The platform authorises, for each feature, which models may be used and which one is used by default.
• Your organisation picks, from that list, the one it prefers for each feature.
• The application applies that choice on every operation. If your organisation hasn't chosen anything, the platform default is used.

In personal scope there is no choice: the platform default is always used.

Where it is chosen

1. Open Organisation > AI models
2. You'll see each application's AI features, the models the platform authorises for each of them and which one is in use
3. Pick the model for the features you want to change
4. Changes take effect immediately for all members of the organisation

Each feature also states where the model in use comes from: your organisation's choice, the platform default, or a feature the platform itself fixes.

Any member can look at the screen; changing the choice is up to the owner and the administrators.

The three degrees of control

Not every feature is decided the same way. Each one carries its degree of control, and the screen says which:

• Your organisation chooses — a person deliberately starts the feature: a chat, a translation, an analysis. This is the common case.
• Your organisation's automations — the feature runs on its own, with nobody in front of it: a screening triggered by an incoming application, a description written when a match appears. There is no one to ask at that moment, so the administrator fixes the model and it applies to every run.
• The platform fixes it — shared infrastructure, or features with a technical requirement of their own, such as reading scanned tender documents in Tender or the automated plugin review in LMS Marketplace. They appear on the screen, so that the answer to "which models does my organisation work with?" is complete, but they cannot be changed.

Long jobs

Some jobs take minutes and produce a single result: writing a course in Creator, translating a file in Translator, analysing the documents of a tender in Tender. In those, the model is fixed when the job starts and does not change midway: a result written with two models isn't one result, it's two.

When a model is retired

Providers retire their models, and the platform retires them from the catalogue behind them. When that happens, the features that had it selected fall back to their feature's default model and keep working: nothing waits for someone to intervene. On AI models you'll see the model now in use and can pick a different one.

You don't find out by chance: when a model your organisation had selected is retired, owners and administrators get a notice, in the app and by email, saying which feature is affected, which model was retired and which one now stands in for it. Notices in this category can be adjusted in the notification preferences.

A job already under way finishes with the model it started with for as long as the provider still accepts it; once it doesn't, the job fails and can be run again.