rglauco keeps the whole stack in-house

rglauco's fork adds two changes that pull in the same direction: privacy and self-hosting. Instead of routing every prompt to a cloud AI provider like Claude or Gemini, Mike can now talk to open models running on your own machine, through tools like Ollama and llama.cpp (both let you run AI models locally rather than calling out to a vendor). Several local models ship as defaults, but you can point it at anything you host.

The second half does the same for files. Documents can now live on local disk as a first-class option instead of a hosted cloud storage service, with short-lived signed links keeping the download experience intact. Cloud storage is still there if you want it - the difference is that you no longer have to.