The architect states her goal.
Knowledge and rules check and enrich the request.
Claude sees only the request, never your data.
The checker reviews it until everything fits.
Back in the model, with a reason for every rule.
Spread across files and heads
Tap a source.
Works with:
Design software: Revit, Archicad, Rhino. AI models: Claude, ChatGPT, Mistral, Kimi.
Three everyday cases. First without pelicad, then with it.
Barracks in modular construction. The module grid and connections are confidential.
Without pelicad the module catalogue goes to the AI provider. Or the AI designs without it and gets it wrong.
With pelicad the catalogue stays in the vault. Claude only gets abstract constraints.
The rules apply. The secret stays in-house.
For the highest requirements, the AI model runs on-premise too.
No answer here? and ask us directly.
MCP is an open standard that lets AI assistants such as Claude or ChatGPT work with other software. Instead of only writing text, the AI can use tools through MCP: read data from a model, create elements or check results. The software provides an MCP server, and the AI app connects to it as a client. Learn more at modelcontextprotocol.io.
For Revit 2027, Autodesk offers the official “Revit Public MCP Server” as a tech preview. You install it as an extension from your Autodesk Account and add it to an MCP-capable app such as Claude Desktop (Autodesk’s announcement). Community servers exist for older Revit versions. pelicad sits between Revit and the AI: every request is enriched with your office standards, and every result is checked before it reaches the model.
Yes. With Archicad 30, Graphisoft offers the “Public Archicad MCP Server” as a beta, according to Graphisoft for the Collaborate and Max subscriptions. The Graphisoft help explains how to connect it to Claude Desktop or ChatGPT. Community solutions based on the Tapir add-on are available too.
McNeel provides the “Rhino MCP Platform” for Rhino 8 and the Rhino 9 beta. You install it from the Rhino PackageManager and connect it to Claude, ChatGPT or other MCP clients (discussion on the McNeel forum).
The vendors’ MCP servers give the AI access to the model. They know nothing about your practice: your naming, details, modules, drawing layouts and approvals. That’s what pelicad adds. Your knowledge stays protected in the vault, skills turn it into rules for the AI, and the checker reviews every result against your standards and building codes.
pelicad is designed to be model-agnostic: Claude, ChatGPT, Mistral, Kimi or a local model, so any AI that can work through MCP or tool calls. Each practice decides which model it uses.
No, not as a whole. Your knowledge stays in the vault. The AI only receives the request and the rules it needs for that specific request, abstracted as far as possible. Whether a provider stores data or uses it for training is governed by agreements such as zero data retention, made with each provider. For the highest requirements, the AI model runs on your own servers too.
The vault is hosted in Europe. If you prefer, you can run it on your own servers, so your office knowledge never leaves the building.
No. The checker is a pre-check against your rules and building codes. It makes designs more robust and saves correction rounds, but it doesn’t replace a permit.
pelicad is being developed together with pilot practices. Click to find out when it’s available for your design software. We’ll then send you the download and installation instructions.
The skill library is free. Vault, learning pipeline and checker are for practices that build their own standard, and we discuss terms in person.
Get the plugin for your design software and design by your own rules.
You’re on the waitlist. We’ll email you as soon as your access is ready.
How would you like to meet?
At BIM World24 & 25 Nov, Munich Online meeting, 25 minwith Yannic Schwarz, co-founder