Memory for agents to remember why, not just what. Notes on the epistemic graph, EP belief propagation, and building agents that reason over what they know.
The origin story behind Tortoise, in three minutes: from Michelin kitchens and Oxford psychology to the graph that computes what a team of agents should believe.
Agent memory needs more than storage and retrieval. Epistemic relationships let agents track why they believe something and propagate changes in evidence.
We're building agents that can write code faster than ever. But writing more code and engineering better systems are fundamentally different problems.
Why recall is the measured part of agent memory and learning is not: the eight memory types, what epistemic memory adds, and how to test whether your agent revises beliefs.
Agent memory isn't a filing problem—it's a reasoning problem. Agents need to track what's true, why we believe it, and how beliefs change with evidence. That requires a world model, not just document storage.
Agent memory is more than a database: it is how information is ingested, stored, and retrieved. A breakdown of the three layers of a memory system, the eight types of agent memory, and how models, harnesses, databases, and memory providers package them together.