Architect builds AI agent that draws and outputs IFC models
Frustrated by repetitive tasks, an architect built Axio Coder, an open-source agent that operates API-less software like MS Paint and exports BIM-ready IFC files.
Stock photo for illustration only, not from the actual event
- An architect built Axio Coder, an AI agent designed to operate computer screens.
- Uses Windows UI Automation instead of pixel guessing to control apps without APIs.
- Operates in a closed loop of generating, rendering, inspecting, and self-correcting.
- Exports industry-standard files like IFC, DXF, and STEP for seamless BIM workflows.
In the current software development landscape, numerous artificial intelligence tools excel at writing and modifying repository files. However, a major hurdle remains unaddressed: interacting with computer programs that lack an application programming interface, such as CAD software, Photoshop, or MS Paint, which architects and engineers rely on daily.
Weary of repeating identical drawings, spreadsheets, and model exports by hand, a practicing architect developed a custom solution named Axio Coder. Built as a single-user desktop application using Electron and Python under an AGPL-3.0 license, the tool is designed to work directly alongside human users within their project environments.
While conventional computer-use approaches rely on taking screenshots and guessing pixel coordinates, this project takes a different route by leveraging the Windows UI Automation tree. Native controls announce their name, type, and value directly to the operating system, functioning similarly to screen readers. This allows the agent to query applications for exact button locations rather than guessing.
Stock photo for illustration only, not from the actual event
A crucial capability of this agent is its vision system. It can examine computer screens, PDF pages, images, and 3D models rendered from any angle, comparing its generated output against reference materials side by side. This establishes a continuous loop of generating, viewing, correcting, and repeating—mirroring human workflow—as demonstrated when it drew a rocket in MS Paint stroke by stroke.
The ability of an AI to interact with software UI elements via accessibility trees rather than raw pixel guesswork represents a significant milestone for automating desktop software without native APIs. By translating visual outputs into structured engineering files like IFC, this approach bridges the gap between experimental AI prototypes and professional architectural workflows.
When provided with a reference such as a photograph, drawing, or dimensioned specification, the agent constructs a parametric model and writes it out in formats widely used across the industry, including IFC, DXF, and STEP. The generated data is measured back against the source formula to ensure geometric accuracy before landing directly inside BIM workflows like ArchiCAD.
Currently operating as a single-user desktop application with minor rough edges, the tool handles complex CAD geometry by backing off when errors occur. The developer has published the source code on GitHub under Axidesk/Axio-Coder and is actively working on enabling the agent to sequence multiple software operations autonomously from end to end.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment