Your source stays where you decide.
LegacyPhoenix runs inside your network. AI is off until you turn it on.
What stays inside, and what can leave.
Text version of the diagram. Inside a box labelled your network: source code, parsers, the map, validation, the new system and the proof. Outside the box, the AI endpoint you choose, connected by one arrow labelled one unit at a time, only when AI is on.
With AI off, the arrow is not there.
Works with AI off
Ingest, the map, validation, reconciliation and comparison are plain code. With AI off you still get the map of your system, the validated model and the proof harness.
When AI is on
In developmentYou choose the endpoint, including one inside a cloud account you already control.
A call carries one unit and the slice of the model around it, never the whole codebase.
Every call is logged with its purpose, the model used and its size, so cost is a number you can see.
An inference is always shown as an inference, and a person can confirm or reject it.
Nothing fetched from outside
What LegacyPhoenix serves and builds comes from your own installation. Components are shipped with the product, with their versions and licences.
The legacy system is observed, not changed
In developmentCapture tools read and record. Test runs go against a private copy, not the live system.