Build agents with human-like memory
We have been obsessing over agent memory because it is the line between chatbots that forget everything and assistants that actually feel persistent and useful. In this workshop, we will walk you through Mastra’s newly improved memory system, why we built it the way we did, and what we learned along the way, including the work that led to an industry-leading high score on LongMemEval.
We will explain the different kinds of memory in Mastra and when to use each one: working memory for tracking user preferences and characteristics, semantic recall for long-term conversation history, and our newly released observational memory. We will share the real tradeoffs, the mistakes we made, and the practical techniques that came out of burning through billions of tokens to get this right.
You will see how to build agents that remember users across sessions, recall relevant context from months of history, and reason correctly about time. We will walk through concrete implementation examples and talk about why RAG is still very much alive for agent memory, how formatting and retrieval choices affect accuracy, and what actually matters for performance and cost in production.
This is a live, hands-on workshop focused on giving you the mental models and patterns you need to build agents with state-of-the-art memory, not just theory.
Hosted by
Alex Booker, Developer Experience at Mastra
Tyler Barnes, Founding Engineer at Mastra
Recording and code examples will be available to everyone who registers.