cad2data: command-line CAD/BIM conversion wrapped in agent instructions and n8n workflows

What cad2data actually solves, and for whom

Construction and BIM teams sit on files that only open in expensive desktop software. A .rvt model needs Revit, a .dwg needs AutoCAD, a .dgn needs MicroStation. The moment you want that geometry and metadata inside a spreadsheet, a database or an LLM prompt, you hit a licensing and automation wall. cad2data targets that wall. The README frames the promise as extraction and transformation with no vendor lock-in and without Autodesk or CAD licenses, driven from DDC UI, command prompts, PowerShell or workflows. The audience is narrow but real: data engineers and BIM coordinators on Windows who want repeatable batch conversion rather than one-off manual exports. The repository description adds the second audience, AI agents, which is why the topics list includes ai-agents, claude-code, opencode and n8n. If your job is moving model data into cost estimation, quantity takeoff or carbon reporting, this is aimed at you. If you are a single architect exporting one drawing a month, the machinery here is heavier than the problem.

The converters are binaries, the repository is glue

This is the first thing to understand and the easiest to miss. The licence badge on the README reads dual: MIT for workflows, scripts and documentation, and proprietary for the binary converters. So the Jupyter Notebooks, the n8n workflow JSON, the PowerShell entry points and the agent instruction files are open. The code that parses a Revit file is not. That split explains the repository layout: a folder named DDC_in_additon holds readme assets, and the substance you clone is orchestration. It also explains why the project can claim no Autodesk licence requirement while shipping closed components. Practically, you are adopting a documented interface to a black box. That is a legitimate architecture, the same shape as many ETL tools that wrap a proprietary engine, but it changes your due diligence. You cannot read the conversion logic, patch it, or audit how a particular Revit category is mapped. You can only test the output against your own models.

Supported formats and where the data lands

Inputs are .rvt, .dwg, .ifc and .dgn, shown on the README badges with an Autodesk logo. Outputs are .xlsx, .csv, .dae, .html, .pdf and .ifc. That output list is the interesting part, because it tells you what the pipeline is for. Spreadsheet and CSV output serve cost estimation and quantity takeoff. HTML output serves the quantity takeoff report workflow, listed in the table of contents as Revit and IFC to HTML Quantity Takeoff. IFC output means the pipeline can act as a format normaliser, turning a Revit or DWG source into an open format that other tools can read. DAE is a mesh interchange format, so geometry export is in scope alongside tabular data. What the material does not state is the fidelity of any of these conversions: which Revit categories survive into .xlsx, whether DWG layers map to columns, or how much metadata is preserved in the IFC round trip. Those are exactly the questions you have to answer with your own files before trusting the pipeline.

Getting it running: CLI first, n8n second

The README's table of contents lists a Command Line Interface section and a Running the Converters section, with the CLI presented under a monitor icon as the primary entry point. The stated execution surfaces are DDC UI, command prompts and PowerShell, all on a local Windows machine. So the minimum viable path is: install the converters, open PowerShell, and invoke the CLI against a file path. The README also points to tutorial videos, including a universal converter overview and a DWG to Excel pipeline walkthrough that uses n8n. Beyond that, the material does not reproduce the exact command strings or flag names, so treat the video walkthroughs and the AI_AGENTS_INSTRUCTIONS folder as the operational documentation rather than assuming a documented flag set. The AI_AGENTS_INSTRUCTIONS folder is described as ready context for AI assistants, which means it contains the prompts and constraints an agent needs to call the converters correctly. That is a genuinely useful artefact: instead of hand-writing tool descriptions for Claude Code or similar agents, you point the agent at a folder the maintainers already wrote for that purpose.

The n8n workflows are the real product surface

Nine workflow families are listed, and they map onto a plausible production sequence rather than a demo reel. Basic conversion covers the four input formats. Advanced settings expose converter options. Batch conversion adds validation and reporting. Multi-format validation checks files before or after conversion. Classification uses AI and RAG to categorise elements. Cost estimation splits into an LLM-driven price pipeline for Revit and IFC and a 4D/5D pipeline built on DDC CWICR. Carbon footprint estimation uses an LLM. A simple ETL workflow feeds LLM use cases. Finally, Revit and IFC to HTML quantity takeoff. Read as a whole, this is a construction data stack expressed as n8n graphs: ingest, validate, classify, price, report. The honest caveat is that n8n workflows are only as portable as their credentials and file paths. Nothing in the supplied material indicates whether these ship with placeholder paths or environment variables, so expect to rewire node parameters before the first successful run.

Where it breaks: Windows-only, binary-only, Revit-version-sensitive

Three constraints stand out. First, the description says local Windows machine, and nothing in the material suggests a Linux or macOS path. If your ETL runs in containers on a Linux cluster, this repository does not fit without a Windows host in the loop. Second, the converters are proprietary binaries, so a security review that requires source access will stop here. Third, CAD conversion is version-sensitive by nature. A .rvt file written by a newer Revit release is a common failure point for any converter, and the material does not publish a supported-version matrix. There is also a scoping question: the repository is described as a workflow for AI agents, but the agents are consumers of the CLI, not part of the conversion engine. If you expected an AI model to interpret drawings, you are looking at the wrong layer. This is a plumbing repository that happens to be agent-friendly.

How it differs from IfcOpenShell and IfcConvert

The obvious open alternative for IFC work is IfcOpenShell, whose IfcConvert binary turns IFC into geometry and tabular formats and is licensed permissively with source available. The difference in approach is architectural, not cosmetic. IfcOpenShell starts from IFC and treats it as the canonical model; it does not read .rvt or .dwg at all, because those are closed formats. cad2data starts from the files you already have, including the closed ones, and normalises outward. That is why it can claim no Autodesk licence requirement while still ingesting Revit files: the proprietary converter does the work that an open library cannot legally or technically do. The trade is transparency for coverage. If your source data is already IFC, IfcOpenShell gives you a readable, patchable, cross-platform toolchain and cad2data adds little. If your source data is .rvt or .dwg, IfcOpenShell is not an option and cad2data is one of the few routes that does not involve buying seats.

Maintenance, licensing and what to check before you commit

The repository is not archived and the last push is dated 2026-06-22, so it is actively maintained, though no releases were retrieved, which means there is no tagged version to pin against. Plan for tracking the main branch. The dual licence is the item that needs a real answer rather than a badge: MIT covers the workflows, scripts and docs, and a proprietary licence covers the binary converters. What that proprietary licence permits for commercial redistribution, internal automation at scale, or bundling into a product is not stated in the supplied material, and the GitHub licence field reads NOASSERTION, which reflects the split rather than resolving it. That is a question for the vendor, not for a licence file. The upgrade cost is mostly re-validation: because you cannot diff the converter internals between versions, every update means re-running your own models through the pipeline and comparing the .xlsx or .csv output against a stored baseline. Budget for that test corpus, because it is the only regression net you will have.

Editorial conclusion

Adopt cad2data if you already run Windows machines, want .rvt, .ifc, .dwg and .dgn files turned into .xlsx, .csv, .dae, .html, .pdf or .ifc without holding Autodesk seats, and are willing to drive it from PowerShell, n8n or an agent that reads the AI_AGENTS_INSTRUCTIONS folder. Do not adopt it if you need Linux or macOS execution, if you cannot accept closed-source converter binaries in your build, or if you expect the repository to contain the conversion logic itself. Before committing, verify three things: which converters ship as binaries and under what terms, whether your Revit versions are covered, and whether the n8n workflows run against your own file shares rather than the sample paths in the repository.

Official sources

1. datadrivenconstruction/cad2data-Revit-IFC-DWG-DGN on GitHub
2. Issues
3. Project website
4. README