Benchmarks

This folder contains the evaluation suites used by EverOS to measure memory quality and agent self-evolution. Use these benchmarks to reproduce reported results, compare memory systems, or evaluate new agent learning methods.

Included Benchmarks

Benchmark	What it measures	Start here
EverMemBench	Long-term memory quality in multi-person group conversations, including factual recall, applied reasoning, and personalized generalization.	EverMemBench/
EvoAgentBench	Agent self-evolution across information retrieval, reasoning, software engineering, code implementation, and knowledge-work tasks.	EvoAgentBench/

How to Use This Folder

• Start with EverMemBench/ to evaluate memory retrieval and answer quality.
• Start with EvoAgentBench/ to evaluate whether agents improve from past experience.
• Use the top-level Benchmarks and Evaluation sections for the project-level benchmark overview.