aisensiy / my-bilibili-rcmd

What is my-bilibili-rcmd?

This is a Chrome extension that watches what you watch on Bilibili, feeds that data to an LLM, and generates a personal interest profile that automatically hides content you don't want to see. It tracks your viewing history—including watch time and completion ratio—then uses AI to figure out what you actually like versus what you've marked as "not interested." The extension then filters your recommendation feed based on that profile, plus any manual keyword rules you add. Everything stays local in your browser; no server, no cloud storage.

Built with TypeScript, React, and Vite. Supports OpenRouter, 智谱 GLM, DeepSeek, and any OpenAI-compatible API endpoint you want to wire up.

Why is it gaining traction?

The hook is privacy-first AI personalization. Most recommendation "optimization" tools send your data somewhere. This one doesn't—your viewing history and API key never leave your machine. The LLM integration is genuinely useful: it reads your watch patterns and generates structured interest tags that you can manually edit. The filtering is immediate and visible—you see exactly why something got hidden.

The multi-provider support is practical. If you're in China, 智谱 or DeepSeek have lower latency. If you want Claude or GPT, OpenRouter has you covered. And if you're self-hosting with Ollama or vLLM, there's a custom endpoint option. One extension, four backends.

Who should use this?

Bilibili power users who are tired of clickbait and algorithm-driven trash in their feed. Developers comfortable configuring an API key who want AI-assisted filtering without surrendering their watch history to a third party. Anyone who watches enough Bilibili to have patterns worth analyzing—tech tutorials, gaming streams, whatever—and wants the algorithm to actually respect their preferences.

Not for casual viewers. You need to set up an API key and understand that the AI only works after you've watched enough videos to establish a pattern.

Verdict

This is a well-architected extension solving a real pain point, but the credibility score of 0.85% and 15 stars reflect a very early-stage project with minimal community validation. The code is clean and the feature set is complete, but there's no test suite visible, the README was unreadable in this review, and you'd be trusting an individual developer with your API credentials. Worth trying if you're technically comfortable and care about local-first privacy—but treat it as experimental software, not a polished product.