Yea, but running local models means high investment at front and less capable models. What I mean by saving cost is more like you can bring your own AI, and we help to make your coding agent work efficiently, like it aware when to verbose and not, which may also helpful to run your local agent faster. In that case, would you think it's helpful?