Show HN: EvalCart – Evals and leaderboard for agent shoppers

Category: ai-ml

Tags: agent-evaluation, benchmarking, shopping-agents, llm-evals, grocery

Score: 5.0/10 (Innovation: 6, Technical: 5, Documentation: 6, Utility: 4)

EvalCart is a benchmarking platform for grocery-shopping agents that provides household tasks, a frozen synthetic catalog, and a deterministic scorer to test whether an agent produces a correct cart (covering items, budget, constraints, pantry avoidance). Its focus on multi-dimensional constraint scoring—including null-scored discretionary choices and separate efficiency reporting—gives it a clear, reproducible way to compare agents. It's a useful niche tool born from a real meal-planning project, though it currently covers only grocery scenarios and has no adoption signal yet.

Target audience: AI/agent developers and researchers building or evaluating shopping agents

Repository: https://www.evalcart.com

View on Hacker News