LoginSign Up
← Back to all sparks
P

Paperless-ngx

COLLAB
Velocity6.3

Document management system that transforms physical documents into a searchable archive.

docs.paperless-ngx.com ↗

Paperless-ngx shipped its 3.0 rewrite, then spent a week putting out the fires.

document-managementself-hostedpaperless-aibreaking-changessearch-backendpatch-cadence
◆Current state
3.0.0 landed on July 22 after a long beta, carrying both the largest feature set in the project's history and nine breaking changes. The five releases since are pure repair: a broken migration in 3.0.1 that forced an immediate 3.0.2, then three patch rounds covering OCR skipping, permission-filtered dedup, Gotenberg conversion, email date parsing, AI suggestion caching and search index edge cases. The pace tells you 3.0 shipped into real deployments fast.
◆Where it's heading
The project has moved from a document scanner-and-tagger to a document platform with AI in the core: Paperless AI, remote OCR via Azure, a document parser plugin framework, tantivy replacing Whoosh for search, file versions and sharelink bundles. The breaking changes are the other half of that story — dropping API v1, Python 3.10, document encryption and the old consumer clears the decks for that platform. Recent patches keep touching LLM plumbing, including passing output language into chat and adding docstrings so the classifier reads better as an LLM tool.
◆Prediction
The 3.0.x patch cadence looks set to run at least another round or two — 3.0.5 is still landing search-compatibility and AI-suggestion fixes rather than tapering. Feature work resuming before that queue empties would be the surprise.

◆Recent moves

  1. 10d ago

    Fifth 3.0 patch round: AI cache keying and search date units

    The fifth repair round since 3.0, and still no new feature work. Two threads stand out: the AI suggestion cache is now keyed by model and endpoint rather than shared across them, and the tantivy backend learned to accept Whoosh-era abbreviated date units so old saved searches keep working. Both are the cost of the 3.0 platform swap being paid down in public.

    View source ↗
  2. 14d ago

    Paperless-ngx v3.0.4

    A fix round covering PDF highlight scrolling, OCR being skipped for tagged PDFs with no real text, permission-visible document dedup under multi-tag filters, and dropdown diacritic normalization. Post-3.0 cleanup, with a couple of LLM-adjacent touches like passing output language into chat.

    View source ↗
  3. 17d ago

    Paperless-ngx v3.0.3

    More 3.0 repair: trusted proxy count for allauth, preserved document fields during Gotenberg PDF conversion, timezone-aware email parsing, and notes without a user no longer breaking the search index. Documentation warns bare-metal upgraders about stale files.

    View source ↗
  4. 19d ago

    Paperless-ngx v3.0.1

    A patch carrying query-performance fixes and a RAG correction that excludes a source document from its own similarity results — but it shipped with a database migration bug and was superseded within the hour. The release notes now tell users to skip it.

    View source ↗
  5. 19d ago

    Paperless-ngx v3.0.2

    A single-commit release fixing the broken migration in 3.0.1. The speed of the turnaround is the story — two days after a major, with a migration that blocked startup.

    View source ↗
  6. 20d ago

    Paperless-ngx v3.0.0

    ⚡ SPARK

    3.0.0 is the release the entire beta line was building toward, pairing the AI, search and plugin work with nine breaking changes that clear out a decade of compatibility. Everything shipped since has been patching it, which is what a major of this size looks like in practice.

    View source ↗