WEAVIATE-SPEC

Weaviate open-source vector DB + Cloud SaaS.

Définition

Weaviate features : (1) Schema-defined collections : objects with properties (similar database tables but vector-aware), class definitions + properties JSON schema. (2) Index : HNSW (default) + flat (small datasets) + dynamic (auto-switch HNSW/flat threshold) + multi-vector. (3) Hybrid Search : combine vector (cosine, dot product, L2) + keyword BM25, configurable alpha weight. (4) Modules : text2vec-openai, text2vec-cohere, text2vec-huggingface, text2vec-jinaai, text2vec-aws Bedrock, text2vec-google PaLM, text2vec-transformers (self-hosted), generative-openai, generative-cohere, generative-anthropic, generative-mistral, qna-openai (RAG built-in answer extraction), img2vec-neural (CLIP vision), multi2vec-clip (multimodal). (5) Query languages : GraphQL (primary) + REST. (6) Multi-tenancy : per-tenant data isolation native. Pricing : OSS free self-hosted ; Weaviate Cloud Services (WCS) pricing $0.05/1M embeddings/month + $0.04/1M queries (low-end). Customers : Stack Overflow, Wallaroo, Instabase, Klarna. $50M Series B 2022.

Origine

Weaviate fondee 2019 par Bob van Luijt + Etienne Dilocker SeMI Technologies ; rebrand Weaviate B.V. 2022 ; Series A $16M 2021 ; Series B $50M 2022 (Index Ventures lead) ; ~100000+ self-hosted instances 2024.

Exemple en contexte

Stack Overflow Teams uses Weaviate self-hosted Kubernetes : ingest ~500K developer Q&A documents embedded via text2vec-openai (text-embedding-3-small), hybrid search GraphQL queries semantic + keyword BM25, developer questions matched relevant existing answers ~50ms latency.

Termes liés

• Qdrant — concurrent OSS Rust.