What it is
feedrater is a sketch, a rough draft, of a local Bluesky feed viewer to help you curate your own feed. It is an experiment in Automated Mindfulness. It adds four extra buttons on every post: thumbs up, thumbs down, a note, and a save. Every rating goes into an SQLite database on your machine.
That database is training data. Over time it builds a picture of your taste. Not what your dopamine-loving brain craves, but what you actually value. The theory is that even as we are being manipulated by our choices of what we read (funny, we are manipulating ourselves for that dopamine hit) that we know how this is making us feel. And maybe if we capture that knowledge we can improve our information diets.
How it works
- Run the app locally. It fetches your Bluesky timeline via the AT Protocol.
- Rate posts as you scroll. 👍 👎 toggle on and off. Add notes. Save things worth returning to.
- Author scores emerge from your ratings. Consistent thumbs up means follow more closely. The opposite means the opposite.
- For authors in the middle, the post text tells you what kind of posts you like from them.
- Eventually: plug that data into a custom Bluesky feed generator. Your taste, compiled, becomes the algorithm.
Why
"There is a clear difference between what I like and what my dopamine-loving brain wants to click on." — David Imel
This is the start of a tool for teaching a machine the difference. Your difference specifically, not the aggregate.
It runs on your machine. Your ratings stay on your machine. Later I might work on a way to run the code on the server while keeping your data on your machine. But that is a tomorrow problem.
Stack
Python · FastAPI · atproto · SQLite · HTMX · Tailwind CDN. No build step. Runs anywhere Python runs.