openbiliclaw is a local-first AI agent that constructs a deep psychological profile of the user from behavior and conversations across connected content platforms, then proactively discovers and recommends content to break filter bubbles. User data is stored in a local SQLite file.
Project overview
Unlike passive collaborative-filtering recommendation systems optimized for platform retention and monetization, openbiliclaw runs locally, stores data in a single SQLite file, and uses psychological bridging logic to actively explore new interest domains the user may have never encountered.
Project type
AI Agent · RAG
Use cases
Knowledge Q&A · Search & Research
Deployment
Refer to project documentation
License
MIT
Best for
Users who want a self-hosted, local-first AI agent that builds a psychological profile to discover content across multiple platforms while keeping all data in a local SQLite file.
General users and developers comfortable running a local backend and configuring an LLM provider and embedding service.
Key capabilities
Analyzes user behavior, feedback, and conversations across connected platforms to infer MBTI, cognitive styles, and deep psychological needs, constructing a five-layer soul profile.
Proactively searches for content the user will like across Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, Reddit, and Bangumi based on their psychological profile.
Actively guesses and explores new domains the user might be interested in but has never encountered, bridging interests with psychological logic to break filter bubbles.
Stores all user data locally in a single SQLite file on the user's disk.
Provides a Chrome/Firefox extension shipped with the application to handle in-platform interactions, collect feedback, and securely pass login sessions to the local backend.
Provides a desktop web interface for large-screen recommendations and a mobile web interface for on-the-go usage.
Allows users to chat with the agent and provide feedback (like, not interested) to continuously refine and change subsequent recommendations.
Limitations and risks
The desktop installer is experimental and unsigned/notarized, requiring manual security bypasses.
Requires logging into platforms via browser extension to initialize the profile for most sources, except Bangumi.
macOS users may encounter 'damaged app' or 'unverified developer' errors requiring terminal commands (xattr) to bypass.
X rate limiting may cause discovery to skip data retrieval.
Getting started
Install the browser extension or download the desktop installer, deploy the backend via binary, Docker, or script, log in to content platforms in the browser, then open http://127.0.0.1:8420/web. Manual configuration of the LLM provider and embedding service is required.
Evidence and sources
GitHub project description: Local-first AI agent that learns who you are, then hunts content you'll love across platforms.