Bouncer: Block "crypto", "rage politics", and more from your X feed using AI

What it is

Bouncer is a browser extension that, it has been reported that, uses AI to filter unwanted posts from your Twitter/X feed in real time. Tell it in plain language what you don't want to see — "crypto", "engagement bait", "rage politics" — and it classifies and hides matching posts as they appear. Relief, at last? For anyone exhausted by repeat hot takes and attention-grabbing noise, Bouncer promises a little peace and quiet in the timeline.

Features

The extension supports multiple AI backends — pick cloud APIs (OpenAI, Google Gemini, Anthropic, OpenRouter), a built-in Imbue backend, or run models locally on your GPU. On-device inference is also supported: it has been reported that local models via WebLLM run entirely in your browser with zero data sent externally. It’s multimodal too — image-aware filtering can flag posts based on pictures, not just text — and there’s reasoning transparency so you can see why a post was filtered. Theme-aware UI, cached results, and a “view filtered” list round out the practical bits.

How it works and how to get it

A MutationObserver watches the feed, the adapter extracts text, images and metadata, and posts are queued to the chosen model which returns a category match and reasoning — matching posts fade out and are added to your filtered list. Results are cached so the same post doesn't trigger repeat inference. Install options include the Chrome Web Store and an iOS app; you can also build from source and enable local or cloud models via API keys. Of course, trade-offs remain: using cloud backends sends data to third parties, while local models require a WebGPU-capable browser and some patience to download and cache model files. Want a cleaner feed, or a new layer of algorithmic gatekeeping? Either way, Bouncer is another sign that personalization — and the desire to escape the firehose — is very much this decade’s digital hygiene trend.

Sources: github.com/imbue-ai, Hacker News