Smart Memory (experimental)

Background Details

Until recently, MemoryPlugin has operated as a simple list. You would add new memories to the list, and the list would be shared with the AI when required to give it access to your memories. While this is very useful, it has limitations.

• Feeding irrelevant info to the AI - Dumping the entire list into a chat would feed tons of irrelevant information to the AI, potentially reducing performance by confusing the AI

• Taking up precious space in the context window - Not every conversation needs the full details of every memory. As you add more memories to MemoryPlugin, the space they take up increases, leaving less space for your actual work with the AI tool. If you have a lot of memories, this could even cause you to run out of the context window altogether

• Costing more - All API based tools charge you by the token, so more info in the context window = more cost. Tools like Claude also set your usage limits based on the tokens you use within a given time period, so this would cause you to hit usage limits faster

To solve these problems and provide you with a better memory experience, we created Smart Memory.

What is Smart Memory?

Smart Memory is a new feature within MemoryPlugin that makes it easier to work with more memories.

Smart Memory works at the Memory Buckets level. It takes all the memories you have within a given bucket, and creates a high level overview of them.

The high level overview is then fed into the AI instead of the full memory text.

The AI is also given details on what additional info the full memories contain, and can request them at any time it considers necessary.