OvermindLab context graph tool
just dropped a tool that actually stitches code to production traces so you can build context graphs. Real user interactions get converted into datasets for training and eval, which means you can compare prompts and models, dig into failures, and fine-tune smaller models on your own product data. The best part is it's open source, supports self-hosting, and you keep ownership of the model weights. repo link in 🧵↓
