Local LLM home agent hits big JSON and routing limits
A user describes a local home automation setup with no cloud service. The system takes a user command, finds one home zone, sends that zone’s device state JSON to a local LLM, and returns device actions. It works for simple zone commands, but fails for global commands like turning off all lights and for questions like asking how many lights are on.
Key points
- The setup uses a local LLM to control home devices.
- The current flow sends one zone’s device state JSON to the model.
- Global commands fail because the code only looks for one zone.
- Question commands need a normal answer, not a device control message.
- The proposed direction is to split commands into zone actions, global actions, and queries.
Quick term guide
- automation
- A way to make repeated work happen without doing every step by hand.
- setup
- The hardware and software arrangement used to make something run.
- cloud
- A remote computer you use over the internet instead of your own device.
- system
- Here, system means a repeatable way to use AI, such as steps, rules, or checks.
- state
- Everything an agent remembers — conversation history, settings, and in-progress tasks
- local LLM
- An AI language model that runs on your own computer instead of on a remote server.
- commands
- Instructions given to a computer or tool to do a specific task.
- compute
- The server power and chips needed to run AI systems.