Strong signal and real results. Worth committing a pilot to.
Weaviate
A mature, self-hostable vector DB with native hybrid search and strong multi-modal support. Ready for production RAG if you can manage the ops.
Open-source·RAG·Infrastructure·DevTool
weaviate.ioOur Take
What It Is
Weaviate is an open-source vector database written in Go that combines vector similarity search with structured filtering and BM25 keyword search in a single query interface. It stores both objects and their vector representations, with 20+ ML model integrations for automatic vectorisation at import time. The GraphQL interface lets you combine semantic search with structured data relationships, and multi-vector embeddings (ColBERT) are now default for all new instances.
Why It Matters
Native hybrid search is Weaviate's standout. Most vector databases treat BM25 keyword search as an add-on; Weaviate builds it in with no extra storage cost. For RAG applications that need both semantic understanding and exact keyword matching (which is most production RAG), this matters. The September 2025 v1.30 release making BlockMax WAND (fast BM25) and ColBERT generally available marked a significant maturity milestone. With 14k+ GitHub stars and 2k+ companies in production, it's proven at scale.
Key Developments
- Dec 2025: v1.35 shipped Object TTL, multimodal embeddings via Weaviate Embeddings, and VoyageAI v3.5/Cohere reranker integrations.
- Nov 2025: v1.34 added 30+ new monitoring metrics for observability.
- Oct 2025: Pricing restructured around three billing dimensions with simplified 14-day sandbox.
- Sep 2025: v1.30 made BlockMax WAND and multi-vector embeddings (ColBERT) GA and default.
- Jul 2025: v1.29 introduced RBAC, async replication, and Weaviate Embeddings GA.
What to Watch
Self-hosting at scale requires Kubernetes expertise, which is the main barrier to adoption. The managed cloud pricing (dimension-based) can get opaque compared to Pinecone's simpler model. Watch whether the enterprise features (RBAC only shipped mid-2025) mature enough to close the gap with Pinecone's managed service. Also track the funding situation: $50M Series B in 2023 with no public follow-on may matter if you're betting on long-term managed service investment.
Strengths
- Native hybrid search: Built-in BM25 alongside vector search with no extra storage cost. Most competitors treat this as an add-on.
- Self-host flexibility: Fully open-source (BSD-3) with Docker/Kubernetes deployment. 14k+ GitHub stars, 2k+ companies in production.
- Multi-modal out of the box: Text, image, and ColBERT multi-vector embeddings natively since v1.30, with integrated vectoriser modules.
- Active release cadence: Seven major releases in 2025, each adding meaningful capabilities.
Considerations
- Operational overhead: Self-hosting at scale requires Kubernetes expertise. Managed cloud pricing can get opaque at scale compared to Pinecone's simpler model.
- Query latency trade-off: 95th percentile at ~45ms for 500K vectors. Competitive but measurably slower than Pinecone's 30ms p99 at 1M vectors.
- Younger managed ecosystem: Weaviate Cloud is younger than Pinecone. Enterprise features like RBAC only shipped mid-2025.
- Funding runway: $50M Series B in 2023 with no public follow-on round. Matters for long-term managed service investment bets.
Resources
Articles
More in Data & Retrieval
Weaviate· Context Engineering· Data Mesh· Embedding Fine-tuning· GraphRAG· Knowledge Graphs· Synthetic Data· Contextual Retrieval· Document Parsing· Pinecone· LlamaIndex· pgvector
Back to AI Radar