11.2 Components and prerequisites
| Component | Role |
|---|---|
| AI model | Multilingual language model (about 57 MB) that turns text into a “meaning” vector. Ships with the server release; identical on server and agents. |
| ada-ml (AI service container) | Reads uploaded example and test documents with the agents’ own text extractors. Internal network only; no internet access. |
| ada-api | Stores examples encrypted, trains, and adds the trained decision layer (vector → probability per type) to the signed policy. |
| Agents | Download the model once, verify it, and score documents locally when a rule needs it. |
Before AI does anything on an endpoint, all of the following must be true:
- The AI model is installed on the server (Overview → Model shows a version and size).
- The AI service is running (Overview → AI service shows Running).
- At least one document type is trained (Overview → Document types shows a number “of your own”).
- A published Classification Rule contains an AI condition for that type.
- The AI classification module is on for the endpoint’s scope (Policy → Rules → Agent Rules → Modules), and the policy is published.
- The endpoint runs a Windows or macOS agent of release 1.3.83 or later.