KaivAIn
KaivAIn reads floor plans the way a draftsman would — vector or scanned. Lines become walls, rooms get areas, and the agent only asks what it can't resolve itself.
What it does
Walls, floors, areas and volumes in seconds. Hatched regions auto-classified as soil, concrete, or fill.
Native PDFs or noisy photocopies — same pipeline. Lines are de-noised and closed into room polygons.
The agent reads the title block, verifies scale, and flags only dimensions it can't resolve itself.
Run it on your own machine. MIT licensed. The pipeline, the agent, the canvas — all yours.
How it works
Drag in a PDF, DWG, or scanned image. Any scale, any resolution. Multi-page PDFs processed page by page.
FLD and DeepLSD detectors find every line. The grouping engine closes walls, identifies rooms, and classifies hatched regions.
DXF with annotated entities, a QTO JSON table, or work live in the canvas editor with the AI agent.
Live demo
01 — raw input
02 — pipeline
03 — extracted data
Ready to start
Self-hosted, open source. No sign-up required.