Most AI assistants start every conversation from zero. No memory of what you discussed last week, no awareness of your priorities, no continuity between sessions. As a Cloud Solution Architect juggling multiple customer engagements, I got tired of explaining myself to a goldfish every morning. So I built one that remembers.

His name is Perry. He's not a product, not a SaaS, not a fancy framework. He's a folder of plain markdown files in a git repo, wired up to GitHub Copilot in agent mode, with a Wiki-style knowledge architecture and a handful of skills that automate the boring parts of my week. He drafts in my voice, challenges my thinking when I'm about to do something dumb, remembers what each customer cares about, and quietly compounds value the longer I use him.

In this session I'll show you how Perry works under the hood, and more importantly, how you could build your own version. We'll cover the architecture (sessions, knowledge, reference, state), the lifecycle that turns a fleeting note into a permanent piece of knowledge, the skills that turn repetitive tasks into one-liners, and the discipline that keeps the whole thing from collapsing under its own weight.

Expect a walkthrough of a real day with Perry: start the session, do some advisory work, checkpoint, end the day, watch the knowledge graph grow. Expect opinions on what makes a good thinking partner versus a glorified autocomplete. And expect a few honest stories about what didn't work, including the parts I had to throw away and rebuild.

You'll leave with a mental model for persistent AI memory, a concrete folder architecture you can copy, and a clear answer to the question: what changes when your AI actually remembers who you are?