MemoryGenesis: 250K Examples for LLM Memory Agent Training

MemoryGenesis is a 250,000-example dataset created by WithinUsAI for training large language models to function as memory-first agents. The dataset is designed to teach LLMs behaviors like capturing, storing, retrieving, and updating information. It was last updated on December 29, 2025.

Use Cases

• Train LLMs to capture new information at runtime based on the described agent behavior.
• Implement durable and ephemeral memory storage with TTL based on the dataset's design.
• Develop retrieval and citation mechanisms for relevant memories using RAG-style behavior.
• Build systems for updating, correcting, merging, and deduplicating memories.
• Enforce privacy and redaction protocols for safe memory handling.

Strengths

• Dataset contains 250,000 examples for training.
• Designed for a specific, advanced training goal: memory-first agent behavior.
• Last update timestamp (2025-12-29) is provided.

Limitations

• Column names and data structure are unknown.
• Row count, file formats, and license details are unspecified.
• Full description requires visiting an external page; metadata completeness is limited.

Provenance

Source

WithinUsAI

Freshness

Last updated 2025-12-29 08:18:17