Bringing AI-assisted circuit design, review, and engineering directly into KiCad.
New here? Read What is the KiCad AI Integration Project? for the project's philosophy, evolution, and long-term vision.
Documentation: Start at the Project Index. User guides: docs/User_Guides/README.md — install, every tab, workflows. Quick try: KiCad Tools → External Plugins → KiCad AI Assistant or python scripts/run_ai_assistant.py --ui.
KiCad AI Integration is an open-source project that integrates modern Large Language Models (LLMs) directly into the KiCad electronic design environment. Rather than a generic AI chat window, it is designed as an AI-assisted electrical engineering reasoning platform that builds structured engineering understanding before asking an LLM to reason about a design.
KiCad is the first host application and reference implementation — not the architectural boundary of the system. The host-agnostic framework stack is AERP (AI-assisted Engineering Reasoning Platform): EKM, AERF, EIE, prompts, providers, and related components. See Platform Architecture and Glossary — AERP.
The project automatically collects engineering context from the active KiCad project—including schematics, PCB layouts, netlists, datasheets, and other metadata—and progressively transforms it into structured knowledge through the Engineering Knowledge Model — EKM and AI Engineering Reasoning Framework — AERF. See Project Overview for the full story of how the project evolved and where it is headed.
The initial implementation targets Anthropic Claude, with a long-term architecture designed to support multiple AI providers through a common abstraction layer.
- Integrate AI directly into KiCad
- Minimize manual copy-and-paste workflows
- Automatically gather engineering context
- Reduce prompt engineering requirements
- Support iterative engineering conversations
- Provide meaningful circuit analysis
- Keep the architecture provider-independent
- Maintain compatibility with future AI models
The KiCad AI Assistant shell provides seven embedded tabs with shared project context:
| Tab | Shortcut | Purpose |
|---|---|---|
| Chat | Ctrl+1 | Multi-turn schematic-aware Q&A with approve-before-send |
| Datasheets | Ctrl+2 | PDF library, attach, AI discovery |
| Simulation | Ctrl+3 | SUBCKT gap scan and spice write-back |
| AERF | Ctrl+4 | Staged analysis (0–7) with EKM write-back |
| Notebook | Ctrl+5 | Engineering Knowledge editor |
| Audits | Ctrl+6 | One-click schematic/PCB reviews |
| Routing | Ctrl+7 | Freerouting autoroute with checkpoint accept/reject |
Launch: KiCad ActionPlugin (Tools → External Plugins → KiCad AI Assistant) or python scripts/run_ai_assistant.py --ui.
See User Guides for step-by-step instructions.
The first release consisted of a Python script that executes within or alongside KiCad:
- Read schematic and PCB from project files
- Construct an optimized AI prompt
- Send requests to Claude via the Anthropic API
- Display responses in wxPython UI
Many Phase 1 goals are now superseded by the unified Assistant shell above.
- True wxAUI dock inside PCB editor (Assistant currently uses a non-modal frame)
- Context preview thumbnail for schematic images
- Clickable component references in AI responses
- Simulation closed loop (ADP-006)
- Deeper PI/SI/EMC guidance beyond current audit templates
- Notebook AI edit proposals
Native KiCad plugin— ActionPlugin shippedEmbedded feature tabs— seven tabs in Assistant shellConversation history—kicad_ai/conversation.jsonper projectMultiple prompt templates— Chat template selectorMulti-provider profiles— Claude + Ollama in Settings- Dockable AI chat window — still deferred (non-modal frame today)
Automated schematic/PCB review— Audits tabStaged AERF with EKM write-back— shipped- Freerouting routing UI — shipped (Routing tab)
- Power integrity / SI / EMC deep analysis — partial via audits + live context
- Component comparison
- Datasheet analysis
- Circuit explanation
- Interactive engineering discussions
- Script generation
- Simulation assistance
KiCad project files
│
▼
Context Collection Engine ──► ProjectContext (DesignSnapshot)
│
├──► Heuristic circuit-family classifier + Circuit Family KB
│
▼
Engineering Inference Engine (EIE)
│
├──► AERF staged analysis (stages 0–7, one LLM call per stage)
├──► Chat (general_review — ad-hoc Q&A)
└──► Simulation / SUBCKT workflows
│
▼
Prompt Builder ──► AI Provider Layer (Claude, …)
│
▼
User approval ──► EKM write-back (curated notebook)
Each AERF stage is an LLM call with deterministic prep (extract, classify, KB excerpts, prior stage JSON). See How AERF Works.
Platform and host detail: docs/Architecture/, Platform Architecture.
KiCad_AI_Integration/
├── README.md
├── PROJECT_OVERVIEW.md # Project philosophy and vision
├── PROJECT_INDEX.md # Primary documentation hub
├── PROJECT_CHARTER.md
├── ARCHITECTURE_DECISIONS.md
├── CHANGELOG.md
├── ENGINEERING_DOCUMENTATION_FRAMEWORK.md
│
├── docs/
│ ├── Architecture/ # System design, ADRs
│ ├── AI/ # AI handbook (Phase 2)
│ ├── Developer_Handbook/ # Setup, environment, integration guides
│ ├── Development/
│ ├── Governance/ # Phase 2 placeholder
│ ├── Specifications/
│ ├── API/
│ ├── Database/
│ ├── Deployment/
│ ├── User_Guides/
│ ├── Reference/
│ └── Templates/
│
├── tasks/ # Implementation tracking
├── archive/ # Retired documentation
│
├── src/
│ ├── context/
│ ├── prompts/
│ ├── providers/
│ ├── platform_core/
│ ├── inference/
│ ├── reasoning/
│ ├── ekm/
│ ├── ui/
│ ├── utils/
│ └── plugin/
│
├── tests/
├── examples/
└── scripts/
See PROJECT_INDEX.md for links to all authoritative documents.
Initial support:
- Anthropic Claude Sonnet 3.5
Planned support:
- Anthropic Claude
- OpenAI GPT
- Google Gemini
- Groq
- Ollama
- DeepSeek
- Additional providers through the provider interface
This project takes security seriously.
Guiding principles include:
- API keys are never hardcoded.
- Credentials are stored securely.
- Projects are never transmitted automatically.
- Users explicitly control what information is sent to cloud providers.
- Support for local AI models will be provided whenever practical.
Phase: Post Track C/D — platform frameworks (Tracks B–D) complete. Phase 1 close-out complete (file-based context, gap-fill, housekeeping). Phase 2 (native plugin, embedded Assistant tabs, multi-turn chat) is the recommended next milestone.
KiCad host (working): Launcher (--ui), schematic context with pin-level connectivity and gap detection, datasheet library and panels (--ui-datasheets), chat UI with Approve & Send and audit templates (--ui-chat), simulation/SUBCKT panel (--ui-simulation), built-in sim model auto-apply, AERF staged analysis (--ui-aerf), Engineering Notebook (--ui-notebook), Claude provider, netlist gap-fill template.
Platform: EKM runtime + CLI (src/ekm/); AERF stage registry, classifier, KB loader, full pipeline, learning loop (src/reasoning/, src/inference/aerf.py); EIE chat and simulation orchestration (src/inference/); EKM write-back from approved AERF stages. Blocking Oscillator reference KB complete (stages 00–07).
See Feature Overview for capability status, platform/host separation, and gaps. To validate against your own project, follow Testing With Your KiCad Project.
Contributions are welcome. See CONTRIBUTING.md for workflow, testing, and platform import boundaries.
MIT License — see LICENSE.
This project builds upon the excellent open-source KiCad ecosystem and modern AI technologies to create a next-generation engineering workflow for electronics designers.
Special thanks to the KiCad development community and the AI research community for making this type of integration possible.