MinusPod is a self-hosted tool that detects and removes ads from podcast episodes. It works by transcribing episodes, using an LLM to identify ad segments from the transcript, then cutting them from the audio and serving a clean RSS feed you can subscribe to in any podcast app. It handles dynamically inserted ads (DAI) using audio fingerprinting and builds a cross-episode pattern library that improves detection over time across episodes and podcast networks. When it gets something wrong, there’s a correction UI that feeds back into the system. Runs in Docker, supports Ollama for fully local inference if you don’t want to use an external API.