#self-hosted

Local Gemma was too slow with AIdaemon until I fixed llama.cpp and the prompt size

I wanted AIdaemon on local Gemma 4 26B through llama.cpp, not Ollama. Generation ran at ~45 tok/s on an M4 Pro. Agent turns still felt stuck because prefill on 14k-token prompts took 8 to 9 seconds before the model wrote a single word.