Skip to content
Schema for serialised Lunr indexes
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

# lunr-schema


JSONSchema definition of the format required for a Lunr index.

This is intended to describe and define the format of a serialised Lunr index. By defining the schema it is hoped that other backends will be able to generate a Lunr compatible schema.


The current expected format for Lunr 2.x is a single JSON object with the following properties:

  • version
  • fields
  • fieldVectors
  • invertedIndex
  • pipeline


Currently this is just a string with the version of Lunr that created the index. This may change to the version of the schema being used.


A list of the names of the indexed fields

Field Vectors

In Lunr documents are split into fields. There will be a vector for every field in every document. Document fields are stored as a vector space, where each token in the index has a specific dimension, and the value of that dimension is the weight of that token in that field/document.

Calculating the weight is probably an implementation detail, but a higher value indicates that the token is more important to that document than a lower weight.

Vectors are represented as a single array. This is specific to the implementation of vectors within Lunr. Both the dimension and values are stored in the same array, though they should be considered a flattened list of tuples.

  [[3, 1.23], [8, 0.98], [17, 3.21]] => [3, 1.23, 8, 0.98, 17, 3.21]

So the even indexed values are the dimensions, and the odd indexed values are the weights.

Inverted Index

A list of tuples of term to index entries. This is a list so that no sorting needs to be done during deserialisation. As such it must be sorted by the term. If not sorted it will cause errors during deserialisation.

Index Entry

An index entry is an JSON object that must contain a property named _index which should be unique for this term. It is the dimension used in the vector space representation.

The other properties should be field names, and the values are posting objects.


A posting is a map from document reference to metadata. By default no metadata is captured during indexing and so the metadata can be an empty object. Otherwise this is a key value store of metadata key to metadata value.


An ordered list of pipeline function names that should be added to the search pipeline. By default this will include just the stemmer as "stemmer".

You can’t perform that action at this time.