CacheWise: Understanding Workloads and Optimizing KVCache Management for Efficiently Serving LLM Coding Agents
For LLM serving systems handling coding agent workloads, CacheWise addresses the specific bottleneck of inefficient KVCache reuse in long-running, tool-calling sessions.
Coding agents, a growing LLM application, exhibit long-running sessions with repeated large prefix reuse, causing KVCache pressure that standard policies handle poorly. CacheWise, a KVCache management layer combining prefix-aware scheduling and reuse-aware eviction, reduces evictions by 2-2.6x and improves session completion time by up to 3.5x.
Coding agents are a fast-growing LLM application, executing as long-running closed-loop sessions in which LLM generations alternate with external tool calls. Yet, unlike chat workloads, their serving behavior has not been studied extensively. We address this gap by collecting a dataset of real-world coding assistant traces. Our analysis shows that coding agent sessions repeatedly reuse large prefixes and create sustained KVCache pressure that conventional LLM serving policies handle poorly. Based on our analysis, we present CacheWise, a KVCache management layer that improves KVCache reuse for coding agent workloads. CacheWise combines prefix-aware scheduling with reuse-aware eviction guided by lightweight predictions from tool call metadata. Implemented in vLLM and evaluated on the collected traces, CacheWise reduces KVCache evictions by up to 2-2.6x and improves total agent session completion time by up to 3.5x.