Lucidos: A Local AI OS That Keeps Your Described Apps Running

I've been building Lucidos, a local AI OS that takes natural-language descriptions of apps and keeps them running on your machine — no cloud dependency, no API keys to manage, no subscription tiers to track.

The core idea is straightforward: you describe what you want an app to do, Lucidos generates the scaffolding, spins up the execution environment, and keeps it alive in the background. Think of it as a lightweight, self-hosted agent layer that bridges the gap between a prompt and a persistent, usable application.

How it works

1. You provide a description — something like "a personal expense tracker that logs entries via CLI and stores them in SQLite."
2. Lucidos parses the intent, selects the right runtime and toolchain, and generates the app skeleton.
3. The app runs locally, persists state, and you can interact with it through the terminal or a lightweight web interface Lucidos spins up automatically.

The whole pipeline is designed to be hands-on and transparent. You can inspect generated code, tweak it, and redeploy without ever leaving your machine.

Why I built it

I got frustrated with AI tools that give you a one-shot answer and disappear. When you need a real, reusable tool — something that runs daily, stores data, or automates a workflow — most current setups fall apart. You end up stitching together APIs, containers, and cron jobs manually. Lucidos handles that glue layer for you.

It's particularly practical for people who want a complete guide to running LLM agents locally without the usual deployment headaches. The step-by-step setup takes minutes, and the beginner-friendly defaults mean you don't need to be a DevOps engineer to get something working.

What it handles today

• CLI utilities (task runners, note-takers, data parsers)
• Simple web dashboards for local data
• Automation scripts that poll, schedule, and respond to triggers
• Multi-step workflows where each step is a distinct tool call

The architecture leans on prompt engineering at its core — the descriptions you write directly shape the generated code, so learning to describe apps precisely pays off fast. I've found that refining a description is often faster than writing the code yourself.

Limitations I'm upfront about

It's not a full-blown operating system in the traditional sense. There's no GUI builder, no app store, and complex multi-service architectures still require manual wiring. The real-world use cases it shines in are personal productivity tools, internal scripts, and rapid prototyping.

I'm sharing this as an open-source project and would love feedback on the architecture and the developer experience. If you've been looking for a practical AI workflow that stays local and persistent, it's worth a look.