Circuitron

Circuitron

**Note:** Circuitron is archived. The project is still usable, but the architecture and dependency model reflect an earlier generation of tooling. See [Project status](#project-status) for details.

Circuitron is an agent-driven PCB design prototyper that converts natural language requirements into SKiDL scripts, KiCad schematics, and PCB layout files. It integrates a multi-agent pipeline with advanced reasoning and retrieval-augmented capabilities.

---


---

Features

• Reasoning-driven workflow – OpenAI Agents coordinate planning, part discovery, code generation, validation, and error correction with transparent chain-of-thought output.
• Retrieval-Augmented Generation (RAG) – A dedicated MCP (Model Context Protocol) server surfaces relevant SKiDL documentation and examples to the LLM for every design step.
• Real part selection – Queries KiCad libraries to ensure chosen components exist and match the design specification.
• Automatic schematic and PCB generation – Produces .sch and .kicad_pcb files alongside netlists (and SVG images of schematics) for direct use in KiCad.
• Iterative correction loop – Agents validate generated code, apply fixes, and run ERC checks until a clean design is produced.
• Containerized toolchain – Uses Docker images for KiCad, the MCP server, and the Python execution environment to guarantee repeatable results.

Architecture Overview

Circuitron orchestrates several specialized agents using the OpenAI Agents SDK. A planning agent first interprets the user prompt and proposes a stepwise design plan. That plan can be edited by the user before execution. Subsequent agents handle part discovery and selection, gather SKiDL documentation via the MCP server, generate code, validate it, and resolve any errors. Output files are produced in native KiCad formats so that engineers can review and finalize designs using their usual toolcha