Comment by ggeorgovassilis

I started using it last week when copilot started acting up against the Deepseek models. It's fast, solid, I love the architecture but be warned: it submits very very many input tokens against the LLM endpoint, so robust & cheap token caching is a must.