Prompt Caching?

[wcc118]

Has anyone considered Open Manus a candidate for prompt caching? I've been using it fairly successfully with a local llm (deepseek-r1-aqwen-14b has performed most reliably thus far) with not very complex tasks. I'm reticent to begin testing with a usage api out of fear of the doom loop. As such I came across prompt-caching in anthropic's documentation, nevermind improving context window performance, a quick look considering how the prompt builds during step execution, this could cut api costs from 20-50% perhaps? Anyhow, I'm going to try to implement but if someone else has looked into it already I'd like to hear about it before spending too much time on it. A quick search of the discussion board here yielded no results.