Free. Local. Yours. BIMChat Is Now Open-Source.

I built BIMChat as an experiment in building with AI and integrating AI into applications and now I want to share it with anyone that might make use of it. It’s a simple interactive web app that allows you to view information about your BIM model in a structured way that doesn’t need to render enormous 3D graphics, doesn’t need an expensive modelling program and can run on even a modestly appointed laptop.

It’s built to make complex BIM data conversational, transparent and shareable, just upload an IFC file, ask questions in plain English, and get instant answers about what’s inside, from how many doors are on the first floor to which materials dominate a particular level. Then output data into pdfs with charts or excel with the raw detail.

What?

At its core, BIMChat turns static IFC models into an easily searchable database.

Once you upload a file, BIMChat:

• Parses the model structure using IfcOpenShell

• Maps and indexes every wall, object, and material

• Builds a searchable database for live queries

• And lets you interrogate it in two modes:

  • Fast Mode: Instant pattern-based responses (no API keys required)

  • Smart Mode: AI-powered deep analysis, if you connect your OpenAI API key

It understands natural language, knows how to map “ground floor” to “Level 0” or “Basement-01,” and can even generate reports - from quick object counts to full Excel exports and compliance summaries.

Everything runs locally, inside Docker. It doesn’t need to connect to OpenAI or other LLM if you want it to be sandboxed, no external data processing, no tracking, no cloud storage.

How?

BIMChat runs as a modular containerised application, built around:

• FastAPI for the backend

• PostgreSQL for data storage

• Redis for caching

• A lightweight web frontend served via Nginx

The Docker setup automatically orchestrates everything: backend, frontend, database and caching layers.

If you’re running on Windows, macOS or Linux, it’s the same three-command process.

From a clean machine, you can be running BIMChat in 15–30 minutes:

# 1. Install Git and Docker

# 2. Clone the repository

git clone https://github.com/ikirugai/BIMChat.git

cd BIMChat

# 3. Start the app

docker-compose up -d

Then visit http://localhost:3003 and enter PIN BIM1.

If you’d like to enable Smart Mode, add your OpenAI API key to backend/.env before launching.

Full setup details are in the Deployment Guide.

Once logged in, try uploading an IFC file - anything from Revit, ArchiCAD, or SketchUp.

You’ll get:

• Hierarchical browsing by object type, family, and instance

• Material-based search (“show me all steel columns”)

• Dynamic floor-level discovery (“chairs on first floor”)

• Dual-mode chat for fast and smart queries

• Hardware catalogue management

• Compliance checking and report exports (Excel/PDF)

Whether you’re checking design intent, validating models, or just exploring, it’s designed to make BIM data easier to question and interpret.

Why?

BIMChat started as an internal testbed, part of my broader work exploring how AI can enhance design, security and compliance. But it’s a genuinely useful tool, and it feels right to open it up. While it’s not a polished SaaS product, it is something that you can run locally and gives you a view of project data that can be tricky to extract from some of the bigger modelling apps.

And like most of my projects, I’m sharing it because collaboration improves it.

If you try it, break it, or think of something it should do I’d love to hear from you. And of course, if you need some help setting it up - ikirugai can help out.

Get started here: github.com/ikirugai/BIMChat