Bouncer

Bouncer is an open-source application and browser extension designed to filter unwanted content from X (formerly known as Twitter) feeds. It enables users to define, in plain language, the types of posts they wish to avoid, allowing for a more personalized and controlled social media experience. By removing posts that match user-specified phrases, topics, tones, or even emojis, Bouncer aims to help users shape their feed to better reflect their preferences.

The tool provides several features to support content filtering. Users can type any phrase, keyword, emoji, or even a color, and Bouncer will remove posts that match those criteria. It also allows users to filter posts by tapping a trashcan icon directly on any unwanted post, after which Bouncer suggests possible reasons for filtering and enables the user to remove similar content with a single click. The filtering process is semantically driven, meaning it can identify relevant posts based on context, not just exact keywords. Bouncer also supports image-aware filtering through multimodal models, enabling classification of posts based on images as well as text. Users can view which posts have been filtered and the reasons for their removal, providing transparency and the ability to adjust filtering criteria as needed. The interface adapts automatically to light, dim, and dark modes.

Bouncer operates locally in the browser, with on-device inference using local models via WebLLM, ensuring that no user data is sent externally. It also supports running models locally on a GPU or utilizing cloud APIs from providers such as OpenAI, Google Gemini, Anthropic, and OpenRouter. 5-4B, which is currently run from the developer's data center, with ongoing work to enable operation directly on user devices. Bouncer does not collect or store identifying data about users.

The tool is available as a Chrome extension and as an iOS app in the Apple App Store, with plans to expand support to Reddit and LinkedIn. Bouncer is free and open source, maintained by a dedicated team focused on transparency and user control.