Skip to content

v4.5.0 | Built-in archive, mobile apps, MCP server

Latest

Choose a tag to compare

@clusterzx clusterzx released this 22 Nov 16:02
c7404c8

Highlights

Built-in archive – Paperless-AI without Paperless-ngx

Paperless-AI can now be your archive on its own: the setup offers the built-in archive next to the companion mode. No Paperless-ngx, Redis or Postgres needed – one container.

  • Getting documents in: upload, a REST API (also the post_document endpoint of Paperless-ngx), watch folders for scanners, IMAP mailboxes with rules, a WebDAV inbox and the share menu of your phone (installable app).
  • Libraries with roles (owner, editor, viewer) for families and teams – every list, statistic and Ask your archive answer only shows what you may see. A default library per user decides where uploads go.
  • Versions with text comparison, a change log with undo for every change, a trash.
  • Full text search built for 200,000+ documents on modest hardware.

Existing installations stay in companion mode – nothing changes until you choose.

Move over from Paperless-ngx

The migration assistant analyses your Paperless-ngx archive, shows who will see what, imports it with the same document ids (AI history, OCR results, calendar and Ask your archive stay valid), keeps it in sync and switches over without a restart – with a rollback. Importing into a built-in archive that is already in use works too. Exports go out again in the Paperless-ngx exporter format or as plain folders.

Workflows

When something happens and the document matches, do something: set tags and fields, extract values with the AI, ask the AI as a condition, send e-mails or webhooks – with test runs, a run log and undo. Prompt rules are now workflows that run before the analysis.

Search syntax and duplicates

  • Search like tag:steuer -tag:erledigt from:adac type:vertrag created:2024-03 title:"Strom 2024" – in both modes, with completion and a help panel.
  • Tags & metadata → Find duplicates suggests tags, correspondents and types that mean the same (case, legal forms, typos, plurals) and merges them in one click.

The mobile apps made for Paperless-ngx

With the built-in archive, Swift Paperless, Paperless Mobile, Paperless Native and Paperparrot connect to Paperless-AI: enter https://your-server/ngx and sign in with your Paperless-AI account (two-factor code included). Browse, search, edit, upload. See Mobile apps.

MCP server for AI agents

Claude, Hermes, OpenCode, Cursor and other agents can now search, read, ask, change and add documents through the MCP server at /mcp. Clients that start a local program (Claude Desktop, Hermes) use the included stdio bridge. See AI agents (MCP). (#969)

API keys with fine-grained permissions

Each key has its own permissions – read, change, delete, manage labels, upload, ask & chat, MCP, webhook, administration – plus libraries, expiry and a rate limit (requests per minute). Starting points like AI agent, reading or Scanner in the key dialog; the MCP configuration is shown right after creating a key.

And more

  • Single sign-on behind a reverse proxy uses the right redirect address (#1012); admin groups and group sync from your identity provider.
  • Custom fields: amounts show as amount and currency; inputs per field type.
  • Tags & metadata can be sorted by name or number of documents.
  • The documentation is split into guides.

Upgrade notes

  • Pull the new image and start it – the database migrations run on start. The search index of Ask your archive moves into data/index.db (a backup goes to data/backups/); with a large archive the first start takes a little longer.
  • Accounts and roles: existing accounts become administrators; accounts added later are members unless you make them administrators.
  • API keys: a key of an administrator acts as administrator only with the Administration permission. Deleting documents and managing labels are permissions of their own now – existing keys that could change documents keep them, new keys need them ticked.
  • Document chat is a tab of every document; the separate page is gone (old links open the tab).
  • Watch folders inside the data directory or the file storage are refused – use a folder of its own such as /consume or /scans.
  • Details: Upgrading.

What's Changed

  • 4.5.0: built-in archive (standalone mode), migration from Paperless-ngx, workflows, search syntax, Paperless-ngx app API, MCP server by @clusterzx in #1024
  • Docs: bring the READMEs and guides up to date for 4.5.0 by @clusterzx in #1025

Full Changelog: v4.4.0...v4.5.0