feat: feat: long-term semantic memory

[mhordynski]

Feature description

Agents should have access to tool / tools giving them capability to save and retrieve relevant chunks of information or facts without necessity of keeping full chat history.

Example architecture may involve VectorStore, Embedding model and Agent.

Agent has one tool or postprocessor which extracts important facts from conversation with the user.

Before each query we do RAG-like query to fetch relevant information. It is important to provide a way to specify a key that will be used to store / retrieve data (for example user_id may be key).

Motivation

Long-term memory allows to further customize agents to meet user expectations.

Additional context

No response