Krypton
Privacy

What Krypton sees of your code

Krypton is built so that scoring never needs your source files. This page says exactly what each way of running it reads, and what it does not.

No source sentScoring works from structure alone.
Names become hashesThe table back to names stays with you.
Check it yourselfkrypton wire prints the exact request.

Two ways to run Krypton

WayWhere your code goesStatus
Command line (krypton)Your machine reads the code and sends the server only a description of its structure. The server never receives your source files. krypton wire prints exactly what would be sent.Available
The hosted demoThe repository you connect is fetched from GitHub by the demo's server with your GitHub access (GitHub authenticates you; Krypton keeps no accounts), read in memory and turned into a syntax tree there; only the tree goes on to the scoring server. Nothing is written to disk, run or kept after the scan, and your GitHub token is forgotten when you sign out or after an hour. The demo does read your code.Live

Two grids as a technical choice

For local CLI analysis, separating syntax structure from line shape into two compact grids is an implementation choice: it preserves the details the scorer needs without sending source text.

What is sent, and what stays on your machine

Sent to the serverStays on your machine
Tree structure and positions of every nodeThe source text itself
Names as keyed hashes, so the server can tell "the same name" from "a different name"The names in readable form, the key, and the map from hash back to name
For expressions it only needs to compare (a returned value, the object a method is called on): a keyed hashThe text of those expressions
Per line: length and where the visible characters areEvery string, number and comment, and the words in them
Which of your files import which, as file pathsThe import statements as written
File path, language and sizeWaiver comments and their reasons (the server learns only that a waiver exists, and a keyed hash of its reason)

What the server sees, and why

Your code's shape. To score code, the server needs its structure: how many functions a file has, how large they are, how deeply they nest and how long the lines are. That is what it scores, and all it needs.

File paths. Folder and file names are sent and appear in the results, so the report can point you to the right file.

Names, keyed. Each name is sent as a keyed hash (HMAC-SHA256). The key is made fresh for every scan, stays in memory on your machine and is thrown away afterwards, so nobody can test a guess against the hashes or link one request to another.