Chroma

Accepts documents or queries, automatically generates embeddings using configurable embedding models (default: all-MiniLM-L6-v2), stores vectors in an in-memory or persistent index, and retrieves semantically similar results ranked by cosine distance. Uses approximate nearest neighbor search (via hnswlib by default) to scale beyond brute-force matching, enabling sub-millisecond retrieval on million-scale collections.

Indexes document text using BM25 (Okapi algorithm) for keyword-based retrieval, enabling fast full-text search without semantic embeddings. Supports boolean operators, phrase queries, and field-specific filtering. Complements vector search by providing exact-match and keyword-proximity capabilities, often combined with semantic search for hybrid retrieval pipelines.

Provides collection-level statistics including document count, embedding count, metadata field cardinality, and index size. Statistics are computed on-demand and can be used for monitoring, capacity planning, and debugging. Supports per-collection metrics without requiring external monitoring infrastructure.

Stores documents as collections with associated metadata (JSON objects), enabling filtering and retrieval based on custom fields. Supports document IDs, text content, embeddings, and arbitrary metadata in a single record. Metadata is indexed and queryable, allowing WHERE-clause filtering before semantic or full-text search, reducing result sets before ranking.

Supports both in-memory (ephemeral) collections for development and testing, and persistent collections backed by SQLite, PostgreSQL, or cloud storage for production use. Collections can be created, queried, and updated with automatic persistence without explicit save operations. Switching between modes requires only configuration changes, not code refactoring.

Exposes Chroma collections as MCP tools, allowing LLM agents and Claude to invoke vector search, full-text search, and document retrieval directly within agentic workflows. Implements MCP resource and tool schemas for semantic search, metadata filtering, and document management, enabling agents to autonomously retrieve context without human intervention or external API calls.

Supports multiple embedding model sources: local sentence-transformers models, OpenAI embeddings API, Hugging Face Inference API, and custom embedding functions. Embedding generation is abstracted behind a provider interface, allowing users to swap models without changing collection code. Embeddings can be pre-computed externally and loaded directly, or generated on-demand during document insertion.

Supports bulk insertion, updating, and deletion of documents in a single operation using upsert semantics (insert if new, update if exists based on document ID). Batch operations are optimized for throughput, reducing per-document overhead compared to individual inserts. Embeddings are generated or updated in batches, leveraging vectorization for faster processing.

Organizes documents into named collections with independent indices, metadata schemas, and embedding configurations. Collections are isolated at the API level, allowing multi-tenant applications to maintain separate document spaces without cross-contamination. Each collection maintains its own vector index, full-text index, and metadata store, enabling per-collection configuration of embedding models and search parameters.

Supports configurable result filtering based on similarity score thresholds and top-k result limits. Queries can specify minimum similarity scores (e.g., cosine distance > 0.7) to exclude low-relevance results, or retrieve only the top N most similar documents. Filtering is applied after ranking, enabling precision-recall tradeoffs without re-running searches.

Supports deduplication of search results based on document ID or metadata fields, preventing duplicate documents from appearing in result sets. Optional re-ranking can be applied post-retrieval using external models or custom scoring functions, enabling multi-stage ranking pipelines (e.g., BM25 first-pass, cross-encoder re-ranking second-pass).