lossless-monorepo/agent-skills/chroma-agent-skills/src/chroma-cloud/templates/schema

Schema

The Schema API configures collections with advanced indexing options, including multiple indexes on the same collection. This enables hybrid search strategies that combine different retrieval methods.

Note: The Schema API is only available on Chroma Cloud.

Why use Schema?

Without Schema, collections have a single dense embedding index. With Schema, you can:

• Add sparse indexes (BM25, SPLADE) alongside dense embeddings for hybrid search
• Configure multiple embedding functions on the same collection
• Fine-tune index parameters for your specific use case

Hybrid search (combining dense and sparse) often outperforms either method alone, especially for queries that mix conceptual meaning with specific keywords.

Important: Schema vs embedding function

When using the Schema API, you cannot pass an embedding function directly to getOrCreateCollection. Instead, you pass the embedding function to schema.indexConfig(). This gives you explicit control over which index uses which embedding.

Imports

{{CODE:imports}}

BM25 sparse index

BM25 is a traditional keyword-based ranking algorithm. It works well when:

• Exact keyword matches are important
• Users search with specific terms they expect to find verbatim
• You want a lightweight sparse index without neural embeddings

BM25 doesn’t understand semantics, so “car” won’t match “automobile”. Use it as a complement to dense embeddings, not a replacement.

{{CODE:bm25}}

SPLADE sparse index

SPLADE (Sparse Lexical and Expansion) is a neural sparse embedding model. It combines the efficiency of sparse retrieval with learned term expansion.

SPLADE vs BM25:

• SPLADE understands synonyms and related terms (like dense embeddings)
• SPLADE produces sparse vectors (efficient like BM25)
• SPLADE generally outperforms BM25 for most use cases
• BM25 is simpler and doesn’t require a neural model

For hybrid search, SPLADE + dense embeddings is typically the best combination. Use BM25 only if you have specific requirements for traditional keyword matching or want to avoid the neural model dependency.

{{CODE:splade}}

Choosing an index strategy