Reproduce “DTOC: Dynamic Tool Output Compression for Adaptive Context Management in AI Agents” 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

Abhay Chaturvedi, Shreya Bhattacharya, Rashmika Gopalkrishnan, Peter van der Putten

DTOC lets coding agents hide selected tool outputs from active context while keeping full results available for later restoration. On DeepSWE, GPT-5.4 completed all project tests on 3 of 6 tasks with DTOC, compared with 2 of 6 without it. Reported input tokens fell from 129,824 to 113,374, and mean steps from 54.2 to 36.7. Results vary by model, and the benchmark covered six tasks with one run per condition.

Summary by alphaXiv.

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