An AI can suggest a wire. This one has to prove the wire belongs there.
atelier-agents is a hardware assistant project that helps produce valid wiring and minimal code from a local database of boards and components.
The problem
General-purpose assistants are useful for explaining concepts, but they become fragile when they need to propose real wiring. In hardware, inventing a pinout or ignoring an electrical constraint can quickly make an answer unusable.
Design & implementation
atelier-agents explores a different approach: combining an LLM runner with a local hardware layer able to validate, canonicalize, and constrain outputs. The project does not rely on text generation alone.
I built a local database of boards and components, explicit validation rules, and a path for producing canonical WIRING_JSON. The system turns that verified context into readable wiring instructions, checks, and minimal Arduino code.
This makes the project interesting not only as a conversational assistant, but as an attempt to make LLM help more reliable in a concrete hardware setting.
What it does
- Hardware assistant specialized in safe wiring guidance
- Local database of boards, parts, and rules
- Deterministic wiring validation
- Minimal Arduino code generation from verified context
Why I built it
atelier-agents aims to reduce the risk of fabricated hardware advice. It shows how LLM generation and technical verification can be combined in a domain where approximation becomes costly very quickly.
