Octavia's Memory

Building Human Memory for AI

Octavia, the Web3 Crypto AI Assistant, is equipped with a sophisticated memory system that allows her to store and recall information just like a human.

Memory Types

This memory system consists of three levels: Local Memory, Global Memory, and Fixed Memory.

Local Memory

Local Memory is Octavia's persistent per-user memory that stores information between user sessions. For example, if a user tells Octavia that their cat's name is George, this information will be stored in Local Memory. This allows Octavia to remember user-specific details and preferences.

Global Memory

Global Memory is where Octavia stores new information that she discovers during her research or learns from external sources. When Octavia performs web research, any relevant information she finds is stored in Global Memory. However, it's important to note that only information fetched externally can go into Octavia's Global Memory to ensure user privacy.

Fixed Memory

Fixed Memory is used to maintain and preserve Octavia's core knowledge about the Octavia project, her personality, and any necessary adjustments to her opinions and thoughts. This ensures consistency in Octavia's behavior and allows her to provide accurate and reliable assistance.

Storage and Recall

We continue to follow the human model as closely as we can in both storage and retrieval.

Storing Memory

Octavia's memory system is designed to mimic how humans store memories. Just like our brains recognize what is important and store it, Octavia's memory system determines the criticality of information and how long it remains relevant. Each memory is assigned to a specific memory type based on the stage of the process:

  • When a user sends a message: Local Memory
  • When Octavia sends a message: Local Memory
  • When Octavia performs web research: Global Memory

This approach ensures that Octavia's memory is organized and relevant to the user's interactions.

Recalling Memory

Octavia's memory recall is based on association, similar to how humans recall memories. Octavia uses a technology called a vector database to associate memories with specific topics. Each memory is plotted in a high-dimensional space, with each dimension representing a different topic or concept. When a user sends a message, Octavia translates it into this high-dimensional space and retrieves the memories that are nearest to the user's message. This allows Octavia to recall relevant information and provide personalized assistance.


In the future, Octavia's memory system will continue to evolve. While there are currently no plans to allow users to delete memories, the team is exploring ways to introduce a natural decay to memories. This means that memories will have a lifespan and can be superseded by new memories over time, just like how human memories fade and are replaced.

Overall, Octavia's memory system enables her to have a human-like memory, allowing her to remember user preferences, store new information, and provide personalized assistance. As Octavia's development progresses, users can expect even more advanced memory capabilities and an enhanced user experience.

Further Reading: