Jan now runs fully on llama.cpp, auto-updates the backend, and lets you switch between llama.cpp builds

Hi, it’s Emre from the Jan team. Jan v0.6.6 is out. Over the past few weeks we've ripped out Cortex, the backend layer on top of llama.cpp. It's finally gone, every local model now runs directly on llama.cpp. We removed Cortex because it was adding an extra hop and maintenance overhead. Folding its logic into Jan cuts latency and makes future mobile / server work simpler. Plus, you can switch to any llama.cpp build under Settings -> Model Providers -> llama.cpp (see the video above). Jan v0.6.6 Highlights: * Hugging Face is integrated in Model Providers. So you can paste your HF token and run models in the cloud via Jan * Jan Hub has been a bit updated for faster model search and less clutter when browsing models * Inline-image support from MCP servers: If an MCP server returns an image (e.g. Jan Search with SERPER API). * *It's an experimental feature, so you need to activate Experimental Features in Settings to see MCP Servers settings.* Please update your Jan or [download the latest](https://jan.ai/). \--- *We've also fixed a bunch of bugs. I'll reply to earlier requests and reports in the previous comments later today.*