Skip to content

Dataset · ZexueHe

Multi-session memory benchmark for AI agents

ZexueHe/memoryarena

Structured agent tasks with subtasks and context for evaluating long-term model memory.

What's inside

A JSONL dataset containing under a thousand records. Each row represents a multi-session task for an AI agent, containing lists of questions, subtask answer lists, and the required background context.

What you can do with it

  • Testing AI agents' ability to retain and use context across sessions
  • Evaluating long-term memory systems and RAG pipelines
Size
up to 1K rows
Downloads this month
23.3K
Likes
18

How to load it

from datasets import load_dataset

ds = load_dataset("ZexueHe/memoryarena")