How it works
One result becomes the next step's input.
That simple rule is the core of Papmil. A workflow is an ordered chain, and a split in the middle can turn one PDF into many without stopping the rest of the sequence.
Run the live workflow- 01
Upload once
The current free proof of concept accepts one PDF up to 15 MB. A workflow record and four ordered steps are created.
- 02
Pass it down the line
Each operation receives the files returned by the previous operation. Resize feeds OCR; OCR feeds the splitter.
- 03
Let the stream branch
The splitter looks for likely document boundaries. One scan can become several PDFs, and every one continues into naming.
- 04
Collect the final documents
The last step returns the filing-ready results to your dashboard. Papmil emails you when the workflow completes.
The actual data path
Hosted, with AI on Papmil's worker.
- Your browserUploads the PDF over HTTPS.
- AWS S3Temporarily stores source, intermediate, and result files.
- Papmil workerRuns Ghostscript, Tesseract, PDF libraries, and Ollama.
- Your dashboardShows final results and the history of every step.
- Automatic cleanupDeletes file copies 30 days after the workflow finishes.
Papmil is not local to your device. Document content is not sent to an external AI API in the current workflow.
Today
One ready-made chain.
The public product currently exposes resize, OCR, semantic splitting, and content-based renaming in that order.
The architecture
No four-step ceiling.
The engine stores operations as ordered data and can keep passing results down a longer chain. The self-service builder is still aspirational.
Explore the aspirational roadmap