I used local models and embedders to find out how coding agents are making decisions for me and how my coding preferences are being saved

I wanted to see if I could take the memories of coding agents like claude code or codex and turn them into visual decision trees that guide me how agents have progressed my codebase and what sort of decisions were made for me at what point of time. This is done with a mix of embeddings generated from analysing the memory files of coding agent tools, they store in different formats for example claude uses .md while codex uses a sqlite instance. This whole tool is open source and runs completely on your machine, check out the repo at https://github.com/pacifio/atlas