Search in Paperless-ngx 3.0: Faster, Smarter, and Highlighted

Full-text search has always been the heart of Paperless-ngx — it’s the difference between an archive and a shoebox. In version 3.0, search got a serious upgrade under the hood, and you’ll feel it the moment you start typing.

A brand-new search engine: Tantivy

The biggest change is invisible but dramatic: Paperless-ngx replaced Whoosh with Tantivy as its search backend. Whoosh is pure Python and, as archives grow, it becomes the thing you wait on. Tantivy is a full-text search engine written in Rust — fast, memory-efficient, and built for this exact job.

All text search now runs through Tantivy, and the indexing side was tuned to match. For users that means snappier queries, quicker indexing after a big import, and search that holds up as your library grows into the tens of thousands of documents.

See your matches: highlighting

Searching is only half the battle; the other half is spotting why a result matched. 3.0 adds search highlighting for titles and content, so your search terms are emphasised right in the result list. No more opening five documents to find the one that actually mentions “dishwasher”.

Friendlier matching for more languages

A search engine is only as good as its matching, and 3.0 fixes several rough edges:

• Diacritics-insensitive filtering. Searching and filtering now ignore accents, so cafe finds café and Muller can find Müller. The UI’s type-as-you-go filtering got the same treatment, with support for matching multiple substrings at once.
• Better CJK support. Chinese, Japanese and Korean title, content and metadata searching was overhauled so those languages index and match correctly — a meaningful fix for multilingual archives.
• Smoother date handling. Date-range queries and year-only searches behave more predictably.

More like this, only better

“More like this” is one of Paperless-ngx’s most underrated features — find one relevant document and jump to similar ones. With the new backend, similarity search is faster and more reliable, and a few panics and edge cases in that path were ironed out during the 3.0 cycle.

The boring part that matters

None of this requires configuration. Whether you run Paperless yourself or use a hosted instance, the search upgrade happens automatically on the 3.0 upgrade — your existing queries keep working, just quicker. If you’ve written advanced Whoosh-style queries, 3.0 restores compatibility for the common ones.

Search is the feature you use every single day, which is exactly why a faster, clearer, more forgiving version of it is the quiet hero of this release. For the headline AI features, see what Paperless AI brings, and for everything else new, check the everyday improvements.