I built a local way to export, search and continue chats from OpenRouter, LM Studio and AI Studio with llama.cpp or OpenRouter

I had a lot of chats in OpenRouter across different models, with basically no proper way to bulk export or search them. So I wrote a js scraper for that. LM Studio was easier because chats are local, while Google AI Studio had the ugliest format. That became ThreadShelf: one local archive with semantic/exact search and MCP. I also had issues with Gemma in LM Studio, so I added a llama.cpp wrapper. Old chats can now be continued either locally through llama.cpp or through OpenRouter. First public OSS release: github.com/ChrystianSchutz/ThreadShelf