Feedpaper

I made this prototype to explore a calmer way to read information feeds. I get a lot of value from seeing other people's work and thoughts on Twitter, but it's also noisy and addictive.

How Feedpaper improves the experience:

1. By default, it only shows you Tweets from yesterday. This is the biggest improvement for me. I can just check it once a day, and there's no incentive to check more often because the list won't change until tomorrow.
2. It groups related Tweets and conversations so they're next to each other in the feed. This makes it easier to read: there's less context switching and you can quickly skip ahead if a topic isn't interesting.
3. Downranks and filters out low relevance and low quality Tweets based on your interests using GPT.

Demo

Intended audience

This is not intended to be a Twitter client, since that would most likely go against Twitter's terms of use. I'm open sourcing it so other people can experiment with different ways to present feeds of information. Some ideas to explore:

• Pull in items from multiple different sources (e.g. newsletters, Reddit, etc).
• Improve the ranking and relevance mechanisms (e.g. by inferring interests and anti-interests, by creating better heuristics for influencer spam, by better merging related items).
• Summarize the feed to create a personalized daily briefing.
• Explore alternate UIs, going beyond a linear feed.

Development setup

After cloning:

Install dependencies:

npm install --prefix=app && npm install --prefix=client

Start development servers:

./dev.sh

Now the Electron app show open. Click "Twitter" in the top tab bar, then log into a Twitter account.

That's it! As long as the window is open, it will periodically refresh to fetch the latest tweets.

You can see the feed by clicking on the "Feedpaper" tab, or going to http://localhost:2346 in your browser.

If you're seeing a 404 page, try restarting the ./dev.sh script.

To enable better grouping of conversations and relevance filtering, click the gear icon and enter your OpenAI API key.

How it works

The code is split into two parts:

• /app: this is an Electron app which does the following:
  • Every 30 minutes, loads Twitter and saves the tweets from the "For You" and "Following" tabs. See app/src/main/main.ts.
  • The tweets are saved in a local SQLite database. See app/src/main/db.ts.
  • If an OpenAI API key is set, it runs an enrichment process on each new tweet to extract entities and score its relevance. See app/src/main/enrichItems.ts.
  • Provides access to the database via a server (http://localhost:2345), which the client (see below) uses. See app/src/main/server.ts.
• /client: this is a Next.js frontend which is where the UI lives.
  • In development, you can access it at http://localhost:2346