Glyph 1.1.9: Local models finally use your GPU on Windows

Windows builds had been shipping a CPU only model runtime, so local models ran on the processor no matter what graphics card was in the machine — when they ran at all, since the llama.cpp runtime download never found a usable release. Both are fixed, and startup is considerably faster besides. Local models now use your GPU on Windows Windows builds previously shipped a CPU only model runtime, so local models ran on the processor no matter what graphics card was installed. They now run on NVIDIA GPUs, covering GeForce RTX 20, 30, 40, and 50 series cards and their workstation equivalents. Fixed model loading failing outright on Windows: the llama.cpp runtime download never found a usable release, so no model could be loaded at all. Fixed GPU detection reporting "Error checking GPU" and 0% usage on Windows 11. Faster startup The backend no longer loads the machine learning stack, the OpenAI client, error reporting, or the HTTP client before it can serve a request. They load on first use instead, which cuts a large amount of work off every launch. Raised the startup budget so the first launch after an update is not cut short while Windows is still scanning the newly installed files. Downloads Model download progress now shows real transferred size, speed, and time remaining instead of "NaN undefined". A model download started during setup now appears in the downloads panel once setup finishes, rather than continuing invisibly. Glyph 1.1.9 is available from fortaify.com, and installed copies update themselves.