Index Types Overview
Schema recognizes six value types, each with associated index types. Without providing a Schema, collections use these built-in defaults:| Config Class | Value Type | Default Behavior | Use Case |
|---|---|---|---|
StringInvertedIndexConfig | string | Enabled for all metadata | Filter on string values |
FtsIndexConfig | string | Enabled for K.DOCUMENT only | Full-text search on documents |
VectorIndexConfig | float_list | Enabled for K.EMBEDDING only | Similarity search on embeddings |
SparseVectorIndexConfig | sparse_vector | Disabled (requires config) | Keyword-based search |
IntInvertedIndexConfig | int_value | Enabled for all metadata | Filter on integer values |
FloatInvertedIndexConfig | float_value | Enabled for all metadata | Filter on float values |
BoolInvertedIndexConfig | boolean | Enabled for all metadata | Filter on boolean values |
Simple Index Configs
These index types have no configuration parameters.FtsIndexConfig
Use Case: Full-text search and regular expression search on documents (e.g.,where(K.DOCUMENT.contains("search term"))).
Limitations: Cannot be deleted. Applies to K.DOCUMENT only.
StringInvertedIndexConfig
Use Case: Exact and prefix string matching on metadata fields (e.g.,where(K("category") == "science")).
IntInvertedIndexConfig
Use Case: Range and equality queries on integer metadata (e.g.,where(K("year") >= 2020)).
FloatInvertedIndexConfig
Use Case: Range and equality queries on float metadata (e.g.,where(K("price") < 99.99)).
BoolInvertedIndexConfig
Use Case: Filtering on boolean metadata (e.g.,where(K("published") == True)).
VectorIndexConfig
Use Case: Semantic similarity search on dense embeddings for finding conceptually similar content. Parameters:| Parameter | Type | Required | Description |
|---|---|---|---|
space | string | No | Distance function: l2 (geometric), ip (inner product), or cosine (angle-based, most common for text). Default: l2 |
embedding_function | EmbeddingFunction | No | Function to auto-generate embeddings from K.DOCUMENT. If not provided, supply embeddings manually |
source_key | string | No | Reserved for future use. Currently always uses K.DOCUMENT |
hnsw | HnswConfig | No | Advanced: HNSW algorithm tuning for single-node deployments |
spann | SpannConfig | No | Advanced: SPANN algorithm tuning (clustering, probing) for Chroma Cloud |
- Cannot be deleted
- Applies to
K.EMBEDDINGonly
SparseVectorIndexConfig
Use Case: Keyword-based search for exact term matching, domain-specific terminology, and technical terms. Ideal for hybrid search when combined with dense embeddings. Parameters:| Parameter | Type | Required | Description |
|---|---|---|---|
source_key | string | No | Field to generate sparse embeddings from. Typically K.DOCUMENT, but can be any text field |
embedding_function | SparseEmbeddingFunction | No | Sparse embedding function (e.g., ChromaCloudSpladeEmbeddingFunction, HuggingFaceSparseEmbeddingFunction, Bm25EmbeddingFunction) |
bm25 | boolean | No | Set to true when using Bm25EmbeddingFunction to enable inverse document frequency (IDF) scaling for queries. Not applicable for SPLADE |
- Must specify a metadata key name (per-key configuration required)
- Sparse vector indices must be declared at collection creation and cannot be added later
- Cannot be deleted once created
Next Steps
- Apply these configurations in Schema Basics
- Set up sparse vector search with sparse vectors and hybrid search