Built a file-native AIOS, running my agency on it, would love feedback

Quick context: I've been building an AIOS and running my own marketing agency on it. It's heavily inspired by ICM, OKF, and the LLM Wiki work, and the through-line from all three is the same. Keep the whole business in plain, structured files instead of a vector DB, so any model can load and reason over it directly, and so the owner keeps their system rather than renting it.

The core is an autonomous agent that does two things:

Onboarding: it runs a conversation that builds your folder structure and context as you talk. No time-consuming manual interview, and you get familiar with the agent from the first message.
Assistant: the same agent then teaches you the system, builds and updates your skills, runs them on demand or on a schedule, and pulls data from your connected tools. Everything is text-first.

Two architecture notes: capabilities are skills you build and run rather than a fixed cast of agents, and it's provider-agnostic so it isn't tied to one model or company.

I built it for my agency first, and I'm now offering it to owners in the 4 verticals I know best (agencies, home services, ecommerce, personal services), improving it with every build.

Where I'd love feedback: the file-native, no-vector-DB bet, the agent-led onboarding as a pattern, and skills-as-capability vs an agent swarm. Where do you think this hits a ceiling? Happy to go deep or show a walkthrough.

www.nuAIOS.com

6 comments

Built a file-native AIOS, running my agency on it, would love feedback