Company backend

[acbits]

Is it possible to use llm as a company backend by sending the current context(line, function, etc.) as the prefix string?

[ahyatt]

I think it might be possible - if the context is short, and the model is small, it could be fast enough to be reasonable. It probably is easy enough to try, but my guess is getting the quality / speed tradeoff right would be a bit tricky.