Reproduce “Aggron: Aggressive Compression of Agentic KV Caches during Head-of-Line Stalls in the Request Queue” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rishabh Ravi, Tony Hong, Chen Wang, Yue Zhu, Mengmei Ye, Hubertus Franke, Fan Lai, Josep Torrellas
Aggron uses admission wait to compress selected history from queued tool-using agent requests, aiming to reduce KV-cache use without putting summaries on the agent’s critical path. It protects some tool outputs and schedules compression separately when the queue head cannot fit. In simulation at 1 program per second, Qwen3.6-35B-A3B has 27% lower P99 latency than Slipstream; this is not a live concurrent-load result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.