DueproofGitHub

Install

With Python

Dueproof needs Python 3.12 or newer and Tesseract.

git clone https://github.com/MaanavKrishna/dueproof && cd dueproof

python3 -m venv .venv && .venv/bin/pip install -e .

On Linux, install the CPU build of PyTorch first to avoid a multi-gigabyte GPU download:

.venv/bin/pip install torch==2.13.0 --index-url https://download.pytorch.org/whl/cpu

Tesseract

System Command
macOS brew install tesseract tesseract-lang
Debian/Ubuntu sudo apt install tesseract-ocr tesseract-ocr-hin (add -tam, -ben, … per language)
Windows the installer from the UB Mannheim build, with the language packs ticked

Only the language packs for the languages you enable are needed. See Languages.

With Docker

docker build -t dueproof --build-arg OCR_PACKS="tesseract-ocr-hin tesseract-ocr-tam" .

docker run -d -p 8000:8000 -v dueproof-data:/data --name dueproof dueproof

docker exec -it dueproof dueproof users add you@example.org --admin

compose.yaml runs Dueproof behind Caddy with automatic HTTPS; see Deployment. The image has not yet been published or tested in CI — build it yourself for now.

Translation weights (optional)

Reading Indian languages works best with the local translation model, IndicTrans2 (AI4Bharat, MIT). The weights are licence-gated, so they are fetched once, by hand:

  1. Accept the licence at https://huggingface.co/ai4bharat/indictrans2-indic-en-dist-200M.
  2. hf auth login with a read token.
  3. dueproof fetch-model

The weights (about 1 GB) land in <data>/models/indictrans2. Every file is checked against a pinned SHA-256 each time it is loaded; the app itself never downloads anything. For an air-gapped server, run step 3 on a connected machine and copy the folder across.

Without the weights Dueproof still works: other-language text is read, searched and shown, and the profile's own-language vocabulary (if any) finds obligations. See Languages.

Check

dueproof doctor

It checks Python, Tesseract and each language pack, the translation weights, the data folder, the database and sign-in, and says what to fix.

Improve this page on GitHub