What we measure
First without AI, with fixed tools:
- A map of your repository: which folders hold code, tests, documentation, configuration or generated files
- Size per language, and how much test code there is next to the code
- The most complex functions, and the files that change most often
- From the git history: how old, how active, and whether most of the work comes from one person
- Runtimes and frameworks that are no longer supported, and whether there are lockfiles
Then an inspection reads the code
An AI model reads through the code, read-only, part by part, starting from the map. It records what it finds as claims, and every claim must point to a file and a line. What can't be checked doesn't go in your view. We are accountable for the check.
Files with secrets, such as .env and keys, are not read by the model.
What you do with it
Your view says for every finding what it means: the risk for the board, the consequences for product and operations, and for developers where it is and what you can do about it. So you see where a clean-up pays off most, and which update can't wait any longer.
Architecture at a high pace
AI does what one change asks. Whether that change fits the architecture, nobody asks, unless someone looks. Continuous checks show where complexity grows and which files keep coming back, before it sets.