Is provider KV caching sufficient for agent swarms and long run agents?

A Hugging Face forum user identified as mhutchw asked whether provider-managed KV caching is sufficient for agent swarms and long-running agents, citing complaints from inference engineers and startups about the lack of manual control over the KV cache. The user said the black-box caching methods of inference providers are particularly frustrating when forking agents from the same cached prefix or storing a cache for a future agent to reuse later, and asked whether the problem is widespread and what solutions exist.

mhutchw https://discuss.huggingface.co/u/mhutchw 1 I’m trying to build a side project in the inference space. I’ve been talking to a few inference engineers and startups and I’ve been hearing how annoying it is to not have manual control over the KV cache at times and just constantly being subject to the black box caching methods of their inference providers. It is particularly annoying for agent swarms when you want to fork agents from the same cached prefix or manually store a cache for a longer period for a future agent to hit later. Im curious if this problem is consistent across multiple people and if there are any solutions for it that people know about.