silentprior/paperless-paddle-ocr

By silentprior

•Updated 6 days ago

PaddleOCR PP-OCRv6 worker for paperless-ngx

Image

Machine learning & AI

Content management system

585

Overview Tags

silentprior/paperless-paddle-ocr repository overview

⁠paperless-paddle-ocr

A CPU-friendly sidecar that OCRs paperless-ngx⁠ documents with PaddleOCR PP-OCRv6 and writes the extracted text back via the paperless-ngx API.

Full source, issue tracker, and docs: https://github.com/silentprior/paperless-paddle-ocr⁠
No poppler / system OCR binaries needed (PDF pages render via PyMuPDF)
Runs as non-root, ships a HEALTHCHECK, multi-arch (linux/amd64, linux/arm64)
Fully configured via environment variables — no image rebuild needed to tune

⁠Supported tags

latest — most recent tagged release
1.0.0, 1.0, 1 — semantic version tags (immutable per-patch tag recommended for production)

⁠Quickstart

docker run -d \
  --name paperless-paddle-ocr \
  -e PAPERLESS_BASE_URL="http://your-paperless-host:8000" \
  -e PAPERLESS_API_TOKEN="your_token_here" \
  -e PAPERLESS_INPUT_TAG="to_ocr" \
  -e PAPERLESS_OUTPUT_TAG="ocr_done" \
  -v paddle-ocr-cache:/app/.paddle_cache \
  -p 8081:8080 \
  silentprior/paperless-paddle-ocr:latest

Or with Docker Compose — see the docker-compose.yml⁠ in the source repo.

Tag any paperless-ngx document with your PAPERLESS_INPUT_TAG (default: to_ocr) and the worker will pick it up on its next poll.

⁠Key environment variables

Variable	Default	Description
`PAPERLESS_BASE_URL`	(required)	URL of your paperless-ngx instance
`PAPERLESS_API_TOKEN`	(required)	Paperless API token
`PAPERLESS_INPUT_TAG`	(none — all docs)	Only process documents with this tag
`PAPERLESS_OUTPUT_TAG`	(none)	Tag applied after a successful OCR
`PAPERLESS_ERROR_TAG`	(none)	Tag applied if OCR fails
`PAPERLESS_PROCESSING_TAG`	`paddle_processing`	Temporary tag applied while a document is being OCR'd
`PAPERLESS_PROCESSED_TAG`	`paddle_processed`	Tag applied after successful OCR to prevent reprocessing
`PAPERLESS_TRACKING_TAG`	`paddle_processed`	Legacy alias for `PAPERLESS_PROCESSED_TAG`
`PAPERLESS_REPROCESS`	`false`	Reprocess completed documents; documents being processed remain excluded
`PAPERLESS_DRY_RUN`	`false`	Log intended changes without writing them
`PAPERLESS_RUN_MODE`	`daemon`	`daemon` (loop) or `oneshot`
`OCR_LANG`	`en`	PaddleOCR language code
`OCR_TIER`	`small`	`tiny` \| `small` \| `medium`
`OCR_ENGINE`	`paddle`	`paddle` \| `onnxruntime` \| `openvino`
`OCR_DPI`	`200`	PDF render DPI
`OCR_THREADS`	`4`	CPU threads for inference

Full reference: see the docstring in ocr_worker.py⁠ or docs/CONFIGURATION.md⁠.

⁠License

MIT — see the source repository⁠ for details.

Tag summary

Recent tags

Content type

Image

Digest

sha256:286a13524…

Size

525.9 MB

Last updated

6 days ago

docker pull silentprior/paperless-paddle-ocr