liveBook · Manning

Working memory

As explained in Domain-Specific Small Language Models

Conversation memory that keeps recent interactions within a session so they can be reused immediately.

Short-term memory tracks the ongoing conversation within a session by keeping recent message history available for immediate reuse. section 14.3

It enriches prompts with prior interactions so a model can narrow queries, improve accuracy, and respond with more relevant context. section 14.3

Typical operational uses here include:

  • Maintaining the latest exchanges in the current session. section 14.3
  • Retrieving recent content to guide the next response. section 14.3
  • Reducing API calls and generated tokens by reusing context. section 14.3

Session history buffer

A memory manager loads recent interactions, slices the last few turns, retrieves other relevant interactions, generates a response, then stores the new interaction with an embedding and extracted concepts. section 14.3

How Domain-Specific Small Language Models covers it

How deep it goes
Defines the session boundary and explains relevance-based retrieval, but stops short of exhaustive algorithm details.
Examples
Uses recent interactions, a memory manager flow, and a stored interaction record with embeddings and concepts.
What it assumes
Assumes basic prompt context and session handling, then distinguishes short-term from long-term memory.
What you can do after
Operate a session memory layer that preserves recent context and reuses it in response generation.
How it connects
Connects to long-term memory, RAG systems, chatbots, AI agents, GraphRAG, embeddings, and spreading activation.

Also explained in

Compare how 7 books explain working memory →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Domain-Specific Small Language Models. The section links point to the parts of the book it draws on.