FAQ

What is this environment?

A self-hosted n8n automation stack, plus a small set of web forms, that lets an administrator or researcher run handwritten text recognition (HTR) preprocessing, inference, and evaluation jobs without writing any code. It came out of the Flow Project; see flow-project.net for background on the wider project. See How n8n fits together for the full technical picture.

Is it free and open source?

Yes, with one licensing nuance worth knowing. This repository, and the preprocessing and inference/evaluation services it calls, are all released under the MIT license; genuinely open source, free to use, modify, and redistribute. See License and Citation on the home page for the details, and how to cite the project.

n8n itself is different: it's source-available under the Sustainable Use License rather than an OSI-approved open source license. It's free to self-host and use internally, including commercially, but you can't resell n8n itself as a competing hosted service. This rarely matters for simply running this environment - it only limits offering n8n itself as a SaaS product built on top of it.

Is this environment still maintained?

The Flow Project, the funded research project that built this environment, concluded in June 2026. There is no dedicated team or funding behind it anymore. Former contributors still check in on issues and pull requests from time to time, on a best-effort basis, but there's no guaranteed response time.

None of that affects whether you can use it: everything here remains free and MIT-licensed (see above), so anyone is free to keep running it, fix things, or take its development further. See Contributing if you'd like to help.

What do I need to run it?

Docker and Docker Compose, at minimum. See Installation for the full requirements and setup steps.

Does this environment train models?

No. It runs preprocessing, inference, and evaluation with an existing TrOCR model. Training happens outside this environment, with whatever training process you prefer. This environment prepares the line-mode dataset a training run would need.

Can I use my own model?

Yes. The inference form accepts any HuggingFace model identifier that follows the TrOCR architecture, including one your team has fine-tuned yourselves. See Choosing a model.

Where does my data actually live?

Datasets pass through this environment but are stored on the HuggingFace Hub. Garage, this environment's own storage, only holds uploaded ZIP files temporarily, for up to seven days. See Data and backups for what n8n itself keeps.

How do I report a bug or ask a question?

Open an issue on the relevant repository: this one for the n8n environment and workflows, or service-trocr-preprocess / service-trocr-inference for the services themselves. If you're not sure which one, this repository is a safe place to start.

How does this compare to other HTR tools?

This environment doesn't do recognition itself - it's an orchestration layer that dispatches jobs to the separate preprocessing and inference services, through n8n's web forms, with no code required to operate it.