AI that shows its working
The investigation layer is an analyst over the record, not a chatbot over the internet. Ask why integration health fell this week, what depends on a system, or what would be affected if it failed, and get an answer built the way a good analyst builds one.
How an investigation runs
- Identify the objects the question is about.
- Read their health and its evidence, walk their dependencies, list what is open on them, note what changed.
- Hand the evidence — and only the evidence — to the model, fenced as data, with every person's name replaced by a token.
- Take the answer back through the grounding check: a fact that cites nothing read is dropped; an answer with none is refused.
- Show facts, inferences (marked, with confidence and what they rest on), uncertainty, next steps, and the full list of what was read.
What the model never gets
Access to the database. Anything the asker's own permissions would not show them. People's names, emails or phone numbers. The chance to state a fact without a citation. The page says it is AI, every time.
Provider-independent, retention-free
The model provider sits behind one interface and can be changed without touching the application. Customer data is sent to a model only under a zero-retention arrangement, and only when the customer has turned the layer on.