Giving AI Agents a Durable Memory

How can AI agents retain project context, working conventions and past decisions across sessions and tools? This talk shows how user- and project-level instructions can route agents through a small, indexed and versioned Markdown knowledge base, giving them the operational context they need without turning every prompt into a briefing document. We will also explore the limits of this approach, how it compares with native memory, and when heavier retrieval systems become a better fit.