Default num_ctx (context window, in tokens) sent in every Ollama
/api/chat request when no operator override is supplied. Ollama’s
server-side default is 2048, which silently truncates prompts; we set
8192 so callers get useful context without per-call configuration.
Default num_predict (max output tokens) sent in every Ollama
/api/chat request when no operator override is supplied. Ollama’s
server-side default is 128, which silently truncates responses.