Feed Filter
A user-controlled AI filter for social media feeds. Developed by Ted Thayer at Pleb Lab, Feed Filter restores "mental sovereignty" by letting users algorithmically control which posts they see, countering engagement-optimized algorithms that prioritize rage and conflict.
The Problem
Social media algorithms are optimized for engagement time, not user wellbeing:
- Facebook reportedly weights rage-bait 5x higher than neutral content
- 1 in 5 people have become estranged from family due to social media-fueled disagreements
- Users have no meaningful control over what appears in their feeds
Core question driving the project:
"What if we could control which posts we see?"
How It Works
- User sets filter rules — Define what you want to see (or not see)
- Network interception — Feed Filter intercepts feed data as it loads (X/Twitter initially)
- Local AI classification — Post data is sent to the nearest server running two VLM (Vision Language Model) instances for nuanced contextual analysis
- DOM rewriting — The feed is dynamically altered in the browser to hide or re-rank posts based on the user's rules
Features
| Feature | Feed Filter |
|---|---|
| Works across sites | ✓ |
| High user control | ✓ |
| Set & forget | ✓ |
| Nuanced contextual understanding | ✓ (via VLMs) |
Architecture
- Browser extension — Primary interface for desktop users
- Mobile app — Companion for phone-based social media
- Distributed VLM servers — Local models classify content with reasoning
- Minimal API dependency — Designed to avoid platform API rate limits and terms-of-service restrictions
Philosophy
Feed Filter is built with sovereign-tech values:
- Minimize reliance on centralized APIs, VCs, and regulatory chokepoints
- Browser and phone as the user's control points
- Local AI where possible to preserve privacy
Roadmap
- Brand and infrastructure established
- Proof of concept built
- Alpha/Beta testing on X/Twitter
- Public Chrome Web Store launch
- Expansion to additional platforms
Relationship to Nostr
Feed Filter aligns with Nostr's values of user control and permissionless publishing. While initially targeting X/Twitter, the architecture could extend to Nostr feeds in the future.
Related
- AI and Bitcoin — Overview of AI and Bitcoin intersection
- Nostr — Decentralized social protocol sharing Feed Filter's values
- PlebDevs — Pleb Lab education initiative (parent org)