OpenBiliClaw: A Local, Private, Self-Evolving Cross-Platform Content Discovery Agent

OpenBiliClaw is an open-source, self-evolving AI agent designed for personalized content discovery across platforms like Bilibili, Xiaohongshu, Douyin, and YouTube. It operates locally and privately, building a deep psychological profile of the user to proactively find content, rather than relying on platform-defined recommendation algorithms.

RDR87Confidence 88%AI AgentContent DiscoveryPersonalizationOpen SourcePrivacyCross-platformRecommendation SystemBilibiliXiaohongshuDouyinYouTube

Why it matters

OpenBiliClaw offers a privacy-first approach to content discovery, giving users more control over their recommendations by operating locally and building a unique psychological profile. This contrasts with traditional platform-based recommendation systems that often prioritize platform objectives and can lead to filter bubbles. Its cross-platform capability also addresses the fragmentation of user interests across different services.

OpenBiliClaw is an open-source, self-evolving AI agent that aims to revolutionize personalized content discovery. Unlike conventional recommendation systems that are controlled by platforms and often lead to content silos, OpenBiliClaw runs entirely locally and privately on the user's machine. It builds a comprehensive psychological profile of the user through usage, feedback, and dialogue, moving beyond simple click-through rates to understand deeper preferences and cognitive styles.

The agent's core innovation lies in its proactive content exploration. Instead of passively matching tags, it actively guesses and seeks out content from various platforms, including Bilibili, Xiaohongshu, Douyin, and YouTube, that might align with the user's inferred interests, even in areas they haven't explicitly explored. This approach is designed to break information cocoons by suggesting content based on psychological bridges (e.g., a mechanical watch enthusiast might enjoy architectural aesthetics).

All user data is stored locally in an SQLite file, ensuring privacy. Users can customize the agent's behavior, including changing the underlying LLM, modifying the psychological profile, or adjusting its 'skills.' The project provides multiple interfaces, including a browser extension for in-platform interactions and cookie synchronization, and desktop/mobile web interfaces for a full recommendation experience. The latest release, `extension-v0.3.61`, focuses on aligning 'watch later' and 'favorites' functionalities across different platforms and improving the browser extension's user experience.