UnbentThe short version
Choose Unbent if
- You want it hosted: nothing to install, update, back up or keep switched on.
- You keep notes with the paperwork, in Markdown, and not only the documents themselves.
- You want it on a phone, in any browser, and working offline.
- You want an AI assistant working in it over the web, with every edit it makes kept in history.
Choose Paperless-ngx if
- You want to run it yourself, free, on hardware you own, with nobody else holding your documents.
- A scanner, a watched folder or a mailbox feeds your archive, and you want rules to run as documents arrive.
- Several people use the archive, with permissions and share links.
- You keep documents, not notes.
Why they differ
Both read a scanned document as it comes in, so that it can be found later by a word inside it. Paperless-ngx has done this for longer and organises more: every document gets a correspondent and a document type, and rules run on each one as it arrives. The differences are where it runs and what sits beside the documents.
Paperless-ngx is software you host yourself: a Docker Compose stack with a database and a broker, on a machine that stays on, reached over your own network, with your documents on your own disk and nobody else's. Unbent is a service: the same reading, done on a server we run, reached from any browser and from a phone that works offline, with written notes and typed fields beside the files, and an export of plain folders, so leaving is as easy as copying a folder. (The same app can also be deployed to a Cloudflare account of your own, on its free tier, if you want the files in your account rather than ours.) Whether you would rather run that machine, or pay someone else to, is most of the decision.
Side by side
| Unbent | Paperless-ngx | |
|---|---|---|
| Price | $60 a year for unlimited notes and 10 GB of files (25 GB more for $12 a year). Two weeks free to try. | Free (GPL-3.0). You run it: Docker Compose with a database and a Redis-compatible broker beside it, on a computer or NAS that stays on. Hosts that run it for you are listed on its wiki, at their own prices. |
| Searching inside files | Note text and the text of every PDF, searched on the server as soon as it arrives, so every device finds it. Photos are read when you press Auto-fill. | Every document is read on arrival (Tesseract, in over a hundred languages; Office files and email through Tika), and search ranks and highlights matches and finds similar documents. PDFs and images, printed text. |
| Fields and tables | Typed fields per notebook (date, amount, select, rating and more), a table view, and filters on any field: receipts by month, warranties by when they run out. | Custom fields of nine types (text, date, money, select, a link to another document and more), plus tags, correspondents, document types and storage paths across the whole archive, and saved views that filter the list by any of them. |
| Offline | In any browser, phone included: your notes, notebooks and tags are kept on the device and searchable there, with the files of the notebooks you choose. | None: it is a web page served by your own machine, reached over your network or a VPN. |
| Encryption | Encrypted in transit and at rest. Not end-to-end: the server reads your notes, so it can search all of them without every device downloading them first. | Plain files on your own disk, behind whatever HTTPS you set up. Nobody else holds them. |
| Export | Everything, any time, from any browser: a folder per note, its Markdown saved exactly as you typed it and its files beside it. Opens with nothing installed. | The exporter writes every original, its PDF/A copy and a manifest of its tags and fields to a folder or zip, from the command line, and the importer reads it back into an empty install of the same version. The files on disk are readable as they are. |
| Apps | A web app for any desktop or phone browser, installable. Android and iOS apps are coming. | A web app that works in a phone’s browser. Phone apps are third-party, listed on its wiki. Nothing to install on a device, and no desktop app. |
| AI and assistants | Auto-fill reads a scan and fills in its title, notebook and fields. An MCP server that Claude, ChatGPT and others connect to over the web, with every edit an assistant makes kept in the note's history. | A classifier trained on your own documents suggests tags, correspondent and type for each new one. Opt-in and off by default, a model of your choosing (OpenAI-compatible or Ollama) suggests titles and dates too and answers questions about documents. A REST API with tokens; no MCP server of its own. |
| Sharing | One person per account. | Several users with permissions per document, and public share links that can expire. |
| Source | Source-available (Elastic License 2.0). You can run your own on Cloudflare's free tier. | Open source (GPL-3.0), on your own machine. |
What Unbent does better
- Nothing to run. Sign in and it is there. No Docker, no database to back up, no updates to apply, no machine that has to stay on, and nothing that stops working while you are away from home.
- Notes beside the documents. A receipt in Unbent is a note: Markdown you can write in, with the scan attached and the date and total in its fields. Paperless-ngx keeps documents, with a note on each; the writing happens somewhere else.
- Notebooks that work like a spreadsheet. A notebook’s fields are its columns: receipts as a table with a total at the bottom, warranties sorted by when they run out. Paperless-ngx filters a list by a field; it does not add one up.
- A phone that works, offline too. Any browser on any phone, installable, with your notes, notebooks and tags kept on the device and the files of the notebooks you choose. Paperless-ngx is a page on your server and has to reach it.
- Assistants over the web. Claude, ChatGPT and others connect with a sign-in and work in your notes from anywhere, with every edit kept in the note’s history and marked as theirs. Paperless-ngx has a REST API, and wiring an assistant to it is yours to do.
What Paperless-ngx does better
- Your hardware, nobody else’s. Free, open source, and every document on a disk you own, behind a door you lock. Unbent is a service: a server we run reads your paperwork, and it costs $60 a year.
- A pipeline. A watched folder a scanner drops into, mailboxes it fetches attachments from on rules, workflows that tag, file, rename and notify as documents arrive, and a PDF/A copy of everything for the long run. Unbent takes files dropped on it, opened with it from a scanner, or sent by an assistant, and files each with one press of Auto-fill.
- Organising the whole archive. Correspondents and document types across every document, nested tags, storage paths that name folders on disk, saved views, and several people with permissions per document and share links. Unbent’s fields belong to a notebook, and an account is one person’s.
Moving between them
Unbent does not read Paperless-ngx's export, and does not need to: the export is your original files. Run the exporter, drop the originals on Unbent's note list, and each becomes a note with its file attached and its text read; a press of Auto-fill fills in the date, total and sender, and an assistant can do that across a notebook. Tags, correspondents and custom fields do not come across.
The other way is as easy, on purpose: Unbent's export is a folder per note with its files beside it, and Paperless-ngx's consume folder takes files as they are.
What this page says about Paperless-ngx was checked in October 2026, against the sources below. Prices and features change; if something here has, we would like to know.
More comparisons
14 days free, no card needed. Then $60 a year.