MOOF Memory Network

Persistent memory is essential for agents that operate over long durations or repeatedly over many sessions. The MOOF Memory Network is a built-in distributed memory store that all agents in a MOOF universe can utilize for storing and recalling information. It combines multiple approaches to support different types of memory:

• Episodic Memory Streams: Agents keep a chronological event log of experiences, including observations, actions, outcomes, and messages.

• Semantic Vector Memory: This offers a vector database storing text embeddings, extending an agent’s context window for unstructured knowledge.

• Structured Memory: In the MOOF Knowledge Graph, Memory Networks store facts structured as triplets or JSON — for example, finalized contract terms.

• Memory Management and Pruning: Memory Networks prevent excessive memory growth by summarizing and compressing outdated information, similar to how agents create high-level reflections from raw data. The Evolution Engine helps determine which memories are valuable for learning.

In practical terms, the Memory Network operates behind the scenes. A user in the Playground might simply tick a box “Enable long-term memory” for an agent, and MOOF handles the rest. Or a template might come pre-configured to log certain interactions.