Reproduce “Readie: Low-Latency Code Sandboxes for 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

Arjun Sivaraman, Zhaoheng Li, Alan Shi, Sanjay Malisetty, Eugene Wu, Yongjoo Park

Readie selects among planned Python memory snapshots to reduce sandbox startup time for agent-written code. It estimates restore time plus imports missing from each snapshot, then chooses the lowest-cost option. On the test set, with a 15 GB budget, initialization fell from 0.98 to 0.33 hours and end-to-end time from 3.88 to 2.96 hours; this comparison uses per-request snapshots with only an 8 GB cache resident.

Summary by alphaXiv.

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