This is a fantastic project! Transitioning from “Vibecoding” (chaotic, prompt-dependent logic) to a structured Finite State Machine (FSM) approach is exactly what the Home Assistant ecosystem needs for professional-grade reliability.
Here is an English version of your post, adapted to maintain your professional yet approachable “Nordic” tone while hitting all the technical high notes for a global audience.
Taming AI Chaos with LIA & the HA-FSM Standard (v0.9.2)
Moin everyone!
I’ve been following the recent discussions around MCP servers and Claude Code. While “Vibecoding” is a blast for rapid prototyping, it’s often a curse for high-availability Smart Homes. We’ve all been there: the AI generates a “working” automation that eventually collapses into a mess of conflicting triggers and delay loops.
To fix this, I’ve been developing smarthomelily—originally a tool for my colleagues to end their YAML chaos, now evolved into a technical framework: the HA-FSM Engineering Standard.
1. The Strategy: FSM over “Vibecoding”
The goal of the standard (currently in v0.9.1 Public Preview) is to move away from unstable “Trigger-Action” piles and toward Finite State Machines (FSM).
- Centralized State: A dedicated
input_selecttracks exactly where the system stands (and survives restarts). - Hardware Realism: Built-in logic for physical constraints like Duty Cycle (DC) and Carrier Sense (LBT)—crucial for Homematic and Zigbee stability.
- Idempotency: Commands are only sent when a state actually changes. This protects your airtime and your nerves.
2. Meet LIA (Lily Intelligence Agent)
To save you from digging through documentation all day, I built LIA. Technically, LIA is a specialized Bash wrapper around Claude Code, optimized specifically for HAOS (Home Assistant OS):
- CLI Power: LIA runs via SSH/Terminal directly on the host (the “final boss” level of integration).
- Embedded Expertise: On startup, LIA automatically loads the FSM standard as context. She doesn’t “guess” your config; she audits your YAML against FIL-Level (Maturity) and flags anti-patterns like hardcoded delays.
- Safe Workflow: Commands like
lia fsm validaterun aha core checkbefore anything is written to your production config.
3. The Vision: The LILY-Agency
This is currently a solo project in the “Roots” phase, but the LILY-Agency is already taking shape in the background. Think of it as an “AI Architecture Firm” for Home Assistant: multiple specialized agents working hand-in-hand, following FSM norms, featuring a graphical drawing board and automated monitoring.
All local, all Open Source.
The project is a work-in-progress (v0.9.1/v0.9.2), so please be kind—it’s still a construction site! If anyone wants to chat about FSM, high availability, or running AI agents in the terminal, let’s connect.
Cheers from Northern Germany,
smarthomelily (Professional Home Automation Patterns)
