Memory

Every new agent session starts from zero. The agent forgets what the user told it yesterday, and stuffing the whole history into the prompt gets slow and expensive.

In this workshop we build our own memory implementation for an agent, without a memory library: what to keep, what to summarize, and what to retrieve.

What we'll cover: - Short-term memory: the conversation and how to keep it within the context window - Summarizing and compacting long histories - Long-term memory: storing facts about the user and retrieving them when relevant - When memory helps and when it makes the agent worse - How to check that memory actually improves answers

You'll leave with an agent that remembers across sessions and a clear idea of which memory approach fits your project.

Who it's for: people building assistants and agents that talk to the same users more than once.

Prerequisites: Python, an LLM API key, basic RAG or agent experience. The code and recording are shared with members after the session.