Imbue, the AI research lab that raised $220 million last year, just released Bouncer. It's a browser extension and iOS app that uses AI to filter posts you don't want to see on X. You describe what bugs you in plain English. "Crypto." "Rage politics." "Engagement bait." The AI reads each post and hides the ones that match. Simple idea, backed by genuine technical work.
The tool supports a range of AI backends. You can run models locally in your browser using WebGPU, which means zero data leaves your machine. Qwen3.5-4B Vision handles local inference and can analyze images, not just text. Or plug in API keys from OpenAI, Google, Anthropic, or OpenRouter for cloud models. Each filtered post comes with an explanation, so you're not operating blind. Results get cached, so the same tweet doesn't trigger another API call.
Imbue has spent years building reasoning-capable agents. A feed filter feels almost mundane by comparison. But it's a practical use of the same technology, and it solves a real problem. Anyone who's opened X lately knows the signal-to-noise ratio is rough.
The Hacker News crowd had the predictable response: just quit social media. Fair. But that's not how most people operate. Bouncer takes a different approach. Meet users where they are and give them control.