Very nice ollama frontend. It automatically recognises the ollama daemon already running on PC if installed. It provides a nice way to interact with local LLMs, complete with web search integration which, having now seen it firsthand, really transforms the usefulness of the models; even those that appear stupid without web search can be good at summarising information, and become actually helpful when they don't have to rely on only their built in knowledge. This extension is probably the easiest way to get any graphical interface for ollama running, particularly with integrated web search.

It does have a few bugs though. Sometimes if you close the window too soon after generating an answer, it won't be saved in your history and you will have to generate it again (usually if you do it before all of the statistics at the bottom become available). Also I have seen clicking the regenerate button make existing answers suddenly disappear (I think after I switched model). Sometimes some questions you asked disappear after a reload even if the answer remains. Another thing is that attaching images and asking vision models about them just results in an error.

I also tried it on my android phone in firefox, connecting to ollama on my laptop, which is recognised by the app to be running. However, on my phone it does not display the drop down menu for selecting a model or prompt, so I cannot use it. It seems that it does not see any models as installed on Android. Do they have to be installed locally on the phone to work?

Overall, has flaws, but is still a fantastic tool, enabling you to put local models to use conveniently instead of just messing around with them in ollama.

Works very well in the current state (02/01/2025). It would be great to have the custom pilot prompts be more configurable (just one "custom" entry isn't a lot) and using the options window creates a 100% cpu utilization for one core (process web extension).
Overall a great addon, very helpful.

Nice extension, but can you fix it to follow different prompt formats. For example, it doesn't respect or use <|end_of_text|> tokens for ChatML models and keeps outputting more and more text.