Bouncer addresses the challenge of managing noisy and unwanted content on social media feeds, specifically Twitter/X. It provides users with a customizable, AI-powered filtering mechanism that operates in real-time directly within the browser. This tool is designed for users seeking a cleaner, more focused social media experience, allowing them to define and filter content based on natural language descriptions of topics they wish to avoid.

How It Works

Bouncer employs a MutationObserver to continuously monitor the Twitter/X feed for new posts. It extracts relevant data, including text, images, and metadata, using a Twitter adapter. These posts are then processed by a user-selected AI model, which classifies them against the defined filter topics. Matching posts are visually hidden with a fade-out animation and logged internally. A caching mechanism prevents redundant inference calls for posts encountered again, and the system offers reasoning transparency, allowing users to understand why specific posts were filtered.

Quick Start & Requirements

• Installation:
  • Chrome/Edge (Web Store): Install directly from the Chrome Web Store.
  • Chrome/Edge (from source): Clone the repository, run npm install, npm run build, then load the Bouncer/ folder as an unpacked extension in chrome://extensions (with Developer mode enabled).
  • iOS: Install from the App Store.
• Prerequisites: A WebGPU-capable browser is required for local model inference. API keys are necessary for cloud-based AI providers (OpenAI, Google Gemini, Anthropic, OpenRouter). Local models are downloaded and cached within the browser.
• Configuration: After installation, access Bouncer's settings to input API keys or enable local models, and select the preferred AI model.

Highlighted Details

• Natural Language Filters: Users can describe unwanted content in plain English (e.g., "crypto," "rage politics").
• Flexible AI Backends: Supports local inference via WebGPU (e.g., Qwen models) and cloud APIs from OpenAI, Google Gemini, Anthropic, and OpenRouter.
• On-Device Inference: Local models run entirely within the browser using WebLLM/WebGPU, ensuring zero data is sent externally.
• Image-Aware Filtering: Multimodal models can classify posts based on their visual content, not solely text.
• Reasoning Transparency: Provides clear explanations for why each post was filtered.
• Theme-Aware UI: Adapts to light, dim, and dark modes.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or project roadmap were found in the provided README.

Licensing & Compatibility

The README does not specify a software license. This lack of explicit licensing information presents a significant caveat for potential adopters, particularly concerning commercial use or integration into closed-source projects.

Limitations & Caveats

The core functionality for local AI inference is dependent on browser support for WebGPU, which may not be universally available. Reliance on third-party cloud AI providers necessitates the management of API keys and potential associated costs. The absence of a stated license requires users to assume non-commercial use or seek clarification before adoption.