Your AI coding agent starts every session like it just met you
I have eight months of Claude Code and Codex logs sitting on my disk. Roughly 1,600 sessions, close to 3 million tokens of me actually working. Every argument about whether something was "done," every time I rejected a design, every "no, do it this way." It is the most honest record that exists of how I think when I build.
And every new session, my agent reads none of it. It starts from zero. Polite, capable, and with no idea who it's talking to.
So I built a small tool to fix that. It's called Ditto, it's one Python file, and it's open source.
Memory is what you told the model. This is what your work proved.
