Fallback active: Using google gemini-2.5-flash (ollama unavailable)

[1Hackoon]

Hello,

Anyone having the same issue? I'm trying to run qwen2.5-coder:7b locally with CPU only. It works on my terminal but when connected to Skales it doesn't work.

When sending a chat I can see Ollama starts ollama ps but Skales says it is unavailable.

[talentsache]

Sorry for the slow reply - Discussions don't notify me reliably.

Quick diagnosis: the message you saw (Fallback active: Using google gemini-2.5-flash (ollama unavailable)) is not actually a bug. It's the Fallback Provider feature working as designed - when your primary (Ollama) doesn't respond, Skales automatically switches to your configured fallback (Gemini) so your chat doesn't die. The wording was confusing though, so in v10.0.4 (shipping today) I've rewritten the banner to make it clearer and added a "Details" button explaining what happened and how to fix it.

For your actual Ollama issue with qwen2.5-coder:7b on CPU: most likely cause is the request timeout being too short for CPU-only inference. First request can take 30-90 seconds as Ollama loads the model into memory. Try: Settings → AI Providers → Ollama → bump the Request Timeout slider to 300s or higher. Also double-check:
curl http://localhost:11434/api/tags

If that responds with your model list, Ollama is fine. Another thing worth knowing: not all Ollama models support tool calling. qwen2.5-coder:7b does, but if Skales detects tool-call failures, it can fall back. Let me know if the timeout bump fixes it.