AI Configuration
The AI assistant is configured once and applies to all modules — EBIOS RM, IEC 62443, Industrial Risks, PIVOT.
Access
The page is only available if the AI option is enabled on your account. Otherwise: "Access to the AI assistant is not enabled on your account. Contact your administrator."
Available engines
| Engine | Call format | Status |
|---|---|---|
| Gemini (Google) | native Gemini | stable |
| Claude (Anthropic) | native Claude | stable |
| Mistral AI | OpenAI | stable |
| Groq | OpenAI | stable |
| HuggingFace | OpenAI | experimental |
| OpenRouter | OpenAI | experimental |
| Ollama (local) | OpenAI | experimental |
| ⚙ Custom | OpenAI | experimental |
Each engine offers a list of models, some of which accept images — useful for pages that analyse a diagram screenshot or a photo taken on site.
Ollama and Custom
Ollama calls a model running on your own machine: no data leaves the workstation. Custom lets you point to a free URL and model, for an internal service compatible with the OpenAI format.
API key storage
Your key stays in your browser
The API key is stored locally only, never sent to the server or saved in the project database. It is obfuscated before storage — this is a protection against accidental reading, not strong encryption. On a shared computer, use a dedicated, revocable key.
What AI does in the application
The assistant steps in at specific points, never in the background:
- AI Scoping — pre-fills workshops from a descriptive text (EBIOS, IEC 62443, RI);
- scenario suggestions (ZCR 2, ZCR 5, PHA, FMEA, HAZOP);
- compliance gap analysis;
- narrative report summaries.
Always to be validated
Each suggestion goes through a preview screen before import. The AI proposes, it never writes directly into the study.