Memlin: Versioned, Collision-aware Memory for AI Agents.

My latest creation, Memlin.ai, is starting its first closed beta with users.

As AI became a bigger part of my daily work, I found myself bouncing between Claude, Cursor, Codex, Anti Gravity, and a growing collection of agents and AI tools.

At first, it was about cost, usage limits, session limits, and moving to another tool to save money, and each tool was better at different things.

I became extremely frustrated carrying knowledge around manually. Every time I switched between computers, tools, or projects, every system wanted its own memory. I was copying context between tools, moving markdown files around, re-explaining decisions, and reloading information that had already been learned. And was paying tokens to process the same knowledge over and over again.

I did some research, and there are open-source projects and commercial solutions out there, but none that I felt were the easy button. I wanted something that just worked. Set it, forget it, and build. (We’re not there yet, but I can tell you it’s already saving me time and hundreds of dollars a week.)

One of my favorite Disney movies growing up was Robots. Most people probably don’t remember it, but one line always stuck with me: “See a need. Fill a need.”

The problem was that knowledge couldn’t easily move between projects, teams, tools, computers, and agents. Every system was learning things, but almost none of them were sharing what they learned.

So I started building something for myself. A way to capture decisions, lessons learned, context, outcomes, and project knowledge once, then make it available wherever it is needed. Not just for the agents, but for people, teams, agents, projects, and whatever comes next.

What started as a side project quickly became something friends wanted to use. Then coworkers. Then people I’d never met.

Somewhere along the way, Memlin.ai was born.

Is it perfect? Not even close. There are bugs, rough edges, and features I already want to redesign. There are probably a few things I’ll discover are terrible ideas the moment real people start using them.

This isn’t a venture-backed startup. No armies of engineers, product managers, QA teams, or customer success departments hiding behind the curtain. It’s something I self-funded, built with the help of AI, and poured an unreasonable number of nights, weekends, and more late-night hours than I’d care to admit into.
It’s still early. Things will break. Bugs will be found. Features will evolve.

The good news is you’ll know exactly who to blame.

If you’re one of the people who has listened to me talk about AI memory, context windows, agents, token costs, or random product ideas over the last few months, congratulations, you accidentally became part of the beta program long before there was a beta program.

Let’s see where this goes.