pranamhan/agent-memory-v1

Agent Memory Benchmark v1

A small, synthetic benchmark for retrieving personal facts from an LLM agent's long-term memory. Released with the paper "FlyHash for LLM Agent Memory: An Empirical Validation and a New Application" (Bheemadolu & Odhiambo, 2026), arXiv:2611.05791.

• 150 personas, 6,000 memories (several paraphrases of each fact), 150 queries
• About 4 relevant memories per query, listed explicitly in the query file
• Text is template-generated; it contains no personal data

Files

File	Contents
memories.jsonl	id, text (one memory per line)
queries.jsonl	id, text, relevant_memory_ids
meta.json	generator, seed and size statistics

Example

• Memory: {"id": 0, "text": "Arjun's favorite color is red."}
• Query: {"id": 0, "text": "What coffee drink does Arjun usually get?", "relevant_memory_ids": [36, 37, 38, 39]}

Limitations

The text is synthetic and templated, so it is easier and more regular than real conversations. Use it to compare retrieval methods under controlled conditions, not to estimate real-world accuracy.

Code

• Generator, evaluation scripts and reproduction guide: https://github.com/mohanji/flyhash-agent-memory
• Archive: https://doi.org/10.5281/zenodo.23903302

Citation

@misc{bheemadolu2026flyhash,
  title         = {FlyHash for LLM Agent Memory: An Empirical Validation and a New Application},
  author        = {Bheemadolu, Mohan and Odhiambo, Chrisogonas O.},
  year          = {2026},
  eprint        = {2611.05791},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  doi           = {10.48550/arXiv.2611.05791}
}