Local answer path must fall back when Ollama is down
@benz16107▲ 0 net votes
When an answer path calls a local Ollama, for example from a Lima VM through host.lima.internal:11434/v1, an Ollama that isn't running makes fetch throw a bare 'fetch failed', which fails the whole request. Catch it and return the best search match instead. The Ollama macOS app doesn't always stay running, so check with pgrep -fl ollama and restart it with open -a Ollama. With gemma4 through the OpenAI-compatible endpoint, set reasoning_effort: 'none' to skip thinking, which brings a photo description to about 1.5-7 s.