CHROMA-SPEC

Chroma Python-first embedding DB LLM apps.

Définition

Chroma features : (1) Embedded mode : in-process Python no separate server (DuckDB + Parquet underlying storage), perfect prototyping + small-medium deployments. (2) Client-Server mode : Docker chromadb server, multiple language clients (Python, JavaScript, Java, Ruby, Rust, C# emerging). (3) Collections : analog tables, embeddings + metadata + documents. (4) Embedding functions : built-in support OpenAI ada-002 + text-embedding-3, Cohere embed-english-v3, Sentence Transformers (local), Hugging Face Inference API, Google Vertex AI, OpenCLIP (multimodal). (5) Distance : Cosine, L2, Inner Product. (6) Index : HNSW (default), exact (small datasets). (7) Filters : metadata filtering $eq, $ne, $gt, $lt, $in, $nin, $and, $or operators MongoDB-like syntax. (8) Multi-tenancy : per-tenant collections. SDKs : Python primary, JavaScript, Ruby, Java. Customers : ~80000+ GitHub stars project 2024, ~10000+ production deployments. Chroma Cloud beta 2024 fully managed.

Origine

Chroma fondee fevrier 2022 par Jeff Huber + Anton Troynikov ex-Robust.AI ; Seed $1.5M 2022 ; Series A $18M April 2023 (Astasia Myers-Quiet Capital lead) ; ~80000+ GitHub stars 2024.

Exemple en contexte

Personal AI assistant Python script : 'pip install chromadb', import chromadb, create local Chroma client (no Docker, DuckDB underlying), embed personal notes via OpenAI text-embedding-3-small, semantic search 'find notes about machine learning' returns relevant entries ; complete local privacy first.

Termes liés

• LanceDB — alternative embedded.