Self-host OpenBiliClaw

OpenBiliClaw is a personal content discovery agent that runs entirely on your own hardware. Unlike platform algorithms that optimize for engagement, it builds a deep psychological profile from your cross-platform behavior, feedback, and conversations, then proactively searches across 14+ sources including Bilibili, Xiaohongshu, Douyin, YouTube, TikTok, X (Twitter), Zhihu, Reddit, Linux.do, Bangumi, V2EX, Weibo, GitHub, Instagram (experimental), and the open web. The backend is a Python 3.11+ FastAPI service with a browser extension (Chrome/Firefox) that handles platform authentication and signal collection. All data — profiles, recommendations, chat history, and embeddings — stays in a local SQLite database. You provide your own LLM API keys (OpenAI, Anthropic, DeepSeek, Ollama, etc.) and can optionally run a local Ollama + bge-m3 embedding model for fully offline operation. The UI is accessible via desktop web (port 8420), mobile web (PWA), a Flutter native app (separate repo), or as a DeepSeek Harness plugin. Installation options include desktop installers, Docker Compose, a one-line install script, or manual source setup. The project is MIT-licensed and actively developed by a solo maintainer.