A podcast that produces itself. An agent writes each episode, voices it locally, and publishes it to a standard RSS feed served from the edge. The episodes explain the very system that made them.
Real audio, generated locally and published through the pipeline below. Press play.
A script becomes speech through a local text-to-speech model. Generating an episode costs effectively nothing.
A Durable Object serializes every write so the feed can never race itself. Reads are served free from the edge.
The feed labels audio by reading its bytes, and supports true HTTP range requests, so scrubbing works in every podcast app.
The result is a clean RSS feed served from an R2 bucket. It validates and adds to any app, at a running cost near zero.
Scan to open the live feed in your phone's podcast app, or copy the RSS URL.