Open ahmedmoorsy opened 3 months ago
The core memories are "always available" vs. being "contextually available". Depending on the prompt & LLM, the memgpt agent may or may not prefer a specific type of memory.
Basically, the core memories are just a simple list of strings for a user (and you could extend this idea to be the agent's own personality traits).
The semantic memories are retrieved based on some relevance criteria, and the way we interface this to the agent may lead it to prefer it
I tried the bot, but I didn’t see it use the
store_core_memory
tool at all. What is its purpose?