Ran into a problem. I was conflating memories, and memory scoring. Memories are things that come from chat conversations. Memory scoring is math related to those memories.
I realize now that I need more than one schema to do it all, which means... time for another separation of concerns! Love those.
Think, a weighting system, summaries of summaries, and metrics on how important each memory is. Tracking things like character development, new skills learned, category of information, and emotional temperature of any given block of conversational items that have been recorded in the memories database.
Scoring across the archive
The scoring system also takes into account pruned context, and sessions that run over multiple context session databases, when factoring in the weight of any group of entries.
This gives us some much needed redundant context that the machine can use to point the relevance of any memory that comes to mind.
But, for now anyway, it's two different processes that do it, until I get the new Node.js based command server running. This makes sense as long as we're talking about back context, which we have plenty of.
I call it, an "identity management system."
The idea being that this combination of tools, essentially, creates a growing persistent identity for your agents, where they're maintaining context, skills, and inside jokes, across potentially gigabytes of user/agent conversation.
I'm excited to see how this changes my agent's behavior as the newer system begins to come online.
If these databases get too much bigger though, I'm probably implementing Postgres or MariaDB.