whiteguo233/OpenBiliClaw

What it solves

OpenBiliClaw is a local, private, and open-source AI Agent designed to break the "information cocoon" of platform-specific recommendation systems. Instead of relying on platform-defined weights and fragmented data across different sites, it creates a unified, cross-platform content discovery experience that prioritizes the user's actual interests over platform retention metrics.

How it works

The system operates as a local backend that integrates with a browser extension to collect user signals and feedback. It builds a deep "soul portrait" (a five-layer psychological profile: events $\rightarrow$ preferences $\rightarrow$ awareness $\rightarrow$ insights $\rightarrow$ soul) to understand the user's personality, MBTI, and cognitive style. Using this understanding, the Agent actively searches for content across multiple platforms—including Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, Reddit, and GitHub—and recommends items with human-like explanations of why they were chosen.

Who it’s for

Users who want a personalized, private content discovery tool that aggregates interests across multiple platforms without giving their data to a central cloud service, and those who wish to escape the narrow loops of traditional collaborative filtering algorithms.

Highlights

• Cross-Platform Integration: Supports a wide array of sources including Bilibili, YouTube, X, Reddit, and more.
• 100% Local Privacy: All behavioral data, profiles, and credentials are stored in a local SQLite database.
• Active Exploration: Uses psychological bridging to guess and suggest new interest areas the user hasn't encountered yet.
• Multi-Interface Access: Available via browser extension, desktop web interface, mobile web, and a native Flutter app.
• Self-Evolving: Continuously refines the user's psychological profile based on likes, dislikes, and direct chat feedback.