When Bas first added me to this group, I could barely have told you what an API was. I've been mostly quiet since, because I was heads down building something, and I finally get to show you. But the win I want to share isn't the one I thought I'd be writing about. A few weeks in, I was at my kitchen table testing a feature, and my own app wrote something back to me. A little message, the kind it sends to a stylist. I read it and I cried. Me. The person who built the thing. It had just made me feel seen, as a hairstylist, and I wasn't ready for that. I'm a solo hairstylist. That's my actual job. The app is called Rooty, a companion for stylists who are tired of being told the only way to stay booked is to become a full time content creator. I built it between clients over about seven weeks. Here's the part I think this group will appreciate, and it genuinely surprised me: the code was not the hard part. The boring receipts: - 212 commits, about seven weeks, built between appointments - One Python file, roughly 8,000 lines, zero dependencies, just the standard library - 132 tests - Multi-tenant, real logins, running live right now with actual stylists signing up - Claude Code did the heavy lifting on the code, Anthropic's Haiku model runs the AI features - I directed all of it. Every product decision, every word of the voice, every call about what it should and shouldn't do. I'm not a programmer. I'm the person who knew what it needed to feel like. That last line is the whole thing, so let me stay on it. The code, honestly, the AI and I got through together. What I could not outsource, what no prompt handed me, was knowing when a message sounded like a brand instead of a friend. When it was technically fine and emotionally wrong. I felt those the way I feel a bad section in a haircut, right away, in my gut. And I sent them back. So the "it feels like it gets me" that people react to, that came from real decisions: - The captions and replies learn from my actual writing, not a formula. They sound like a person because they're trained on one. - Every AI feature is its own little room with exactly one job. None of them is a general chatbot. A tool that tries to do everything feels like nothing. A tool that does one thing warmly feels like a friend. - And me. I was the reject button. I read hundreds of generated lines and killed every one that rang false.

I’ve been building our agency operating system inside Claude Code, and memory quickly became one of the harder architectural problems. Most AI memory systems use a global store with similarity-based retrieval. That can work for a single project, but it becomes risky when the same system handles multiple clients. Relevant context for one client should never be retrieved while working on another. So I built Agency Memory Kit around a different model: Each client folder is its own source of truth. The core mechanics are intentionally simple: - Context loads in a fixed, task-specific order instead of relying on fuzzy retrieval. - New learnings are written back to the relevant client folder. - Ambiguous learnings go into quarantine instead of being assigned automatically. - A weekly process deduplicates memory, recovers useful learnings from missed sessions, and extracts cross-client patterns without carrying over client-specific details. - Recurring tool mistakes can become proposed guardrails, but promotion, archiving, and hard blocks always require human approval. The plugin is the engine. Your data stays in plain Markdown files inside your own folder structure, where it can be reviewed, edited, and versioned with Git. The daily hooks run locally and send no telemetry. The optional weekly consolidation uses your own Anthropic API key. Agency Memory Kit is currently on v0.2.10. It’s open beta, MIT licensed, and tested end to end on macOS and Windows. If you’re building a multi-client or multi-project Claude Code setup, try it! I'm interested in what you would change. Repo here: https://github.com/krsnczky/agency-memory-kit