Articles

Indx 6 is here, in beta

Word indexing with up to three typos, ranking you control field by field, and faster, leaner search. IndxSearchLib 6.0 and the Indx server 2.0 are both in beta.

Update

Indx 6 is the biggest change to how Indx searches since it began. Words are now indexed as words, a misspelled word is looked up within one, two or three edits, and the pattern matching Indx is known for is still there, chosen field by field. Searches are faster and use less memory, and you decide more of the ranking yourself.

It ships as two betas: IndxSearchLib 6.0, the engine on NuGet, and the Indx server 2.0, which runs on it.

Results people see as relevant, not only ranked correctly.

Finds what people mean

  • Word indexing. A field's words are indexed as words. A query word is looked up as typed, then within one, two or three edits, and a record holding the words as typed is never placed below one holding a misspelling.
  • Pattern matching, field by field. Matching inside words, with its tolerance for heavily misspelled text, is still there. A field gets word matching, pattern matching, or both, each with its own typo tolerance and its own weight.
  • Compound words, found by their parts. "Brun ost" finds "brunost", and a compound finds the records holding its parts. It works from the words in your index, with no dictionary and no language setting.
  • Synonyms that widen. With film and movie as synonyms, "war film" also finds the war movies, and the records holding the typed words come first.

Ranking you control

  • Fetching and ranking, apart. How records are fetched and whether they are reranked are two choices. The reranker, called Coverage in Indx 5, orders the records by how well they cover the query. It is on by default and fast enough to leave on.
  • One count for everything. The count, the facets and the list come from the same set of records, so the number beside the list and the numbers in the filters always agree.
  • Bury. Take the records a filter holds down the list, discontinued products say, and move nothing else.
  • Filter expressions. Write a whole condition as one: brand IN ["nike", "puma"] AND price BETWEEN 100 AND 2000. New filters for a set of values, a distance from a position, and dates.
  • A match report. Ask, and each result says how every query word was met, and in which field.

Faster and leaner

Searches take less time and allocate less memory than in Indx 5, most of all on datasets with many searched fields or long texts. Loading is faster too, and so are facets over large result sets and searches with many filters. Vector search has a new graph: higher recall, quicker to build, and less memory for each document with a vector.

The Indx server 2.0

The server moves to Indx 6 with everything above, and its console grows with it:

  • Records. Each dataset's records as a table, searched as you type, filtered and ordered, with a row that shows how each word matched. A scan reports the records that lack a value or have text cut at indexing.
  • Journeys in Statistics. Where visitors go after a search, a facet value or a catalogue page: to a chosen result, a conversion, or nothing.
  • Search or ask (experimental) tells a question from a search more surely, and says whether what is asked about is in the results.
  • Observr (Pro and Enterprise, early access) reads your statistics each morning and proposes what to change, and answers questions about your datasets on demand. Nothing changes until you say yes.

Upgrading from Indx 5

Indx 6 is a breaking release on purpose. Most code written for 5.0 needs changes before it builds, and result lists change by design, so judge relevance again after upgrading rather than comparing lists with Indx 5's. The migration guide says where every 5.0 setting went. For React, @indxsearch/intrface and @indxsearch/indx-types 4.0 are the clients for the new server.

Try it

IndxSearchLib 6.0 is on NuGet as a beta:

dotnet add package IndxSearchLib --prerelease

The Indx server 2.0 is open source on GitHub, to run on your own machines or in your own Azure subscription.

Both are betas, so your feedback shapes them. Tell us what works and what does not.

More articles

Let your coding agent integrate Indx

A skill, agent-readable docs, and a built-in MCP server. Describe what you want searchable, and your agent does the rest. Updated for Indx 6.

Indx Core v5 is launched, and Indx v2 is in beta

A faster search engine, now released, and a redesigned server in beta. See what is new and how to try them. Updated October 2026 with Search or ask, Observr and search statistics.