Reproduce “Re-Reading Is Not Free: Token Traces and Cache-Aware Energy Estimation for LLM Agents in Software Engineering” 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

Tina Vartziotis, Ketevan Peranidze, Rodopi Kosteli, Ezgi Sarikayak, George Dasoulas, Michael Keckeisen, Sotirios Kotsopoulos, Francesca Dominici

Authors measure accumulated context and energy in a coding-agent workflow across 42 sessions and 752 task episodes. They adapt an energy model to distinguish fresh-token prefill from attention reads of cached context. For 3,829 requests, the hybrid estimate is 248 kWh versus 9.89 kWh for EcoLogits; neither is validated against metered energy for cached contexts.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.