A local reading studio that turns PDFs, Markdown, and Obsidian exports into spoken audio — stripping citations, headers, and figure junk first. The rewrite engine (Gemini or Ollama) and the voice engine (Edge, Kokoro, Piper) are separate controls, and it exports an MP3 plus a listening script, with audiobook export when ffmpeg is around.
The hard part isn’t the plumbing — it’s output trust. The audio only matters if it feels faithful, natural, and low-friction.