molt is an open source coding agent for the terminal and desktop. It runs any OpenAI compatible model, plus Anthropic. When the model says it is done, molt runs the project's .molt/done.yml against the real disk. Failures go back to the model. Accepts and refusals both get a receipt.
I built molt to make “done” mean something you can check. Instead of trusting a model's claim, it verifies the project on disk and sends failures back for another attempt, while keeping a receipt for both accepts and refusals.