# Cloudflare Vectorize [[Cloudflare]] Vectorize is a globally-distributed vector database for embeddings, designed to pair with [[Cloudflare Workers AI]] for RAG (Retrieval-Augmented Generation) and semantic search workloads. Indexes store dense vectors with optional metadata; queries return the top-K nearest neighbors via cosine, Euclidean, or dot-product distance. It's the "V" in Cloudflare's AI stack: ingest text → embed with Workers AI → store in Vectorize → query at request time → augment LLM prompts. Same edge-native runtime characteristics as the rest of the platform: bound directly from a Worker, no separate connection pool, no cold starts. ## Why It Matters Building RAG without a managed vector DB means running Pinecone, Weaviate, Qdrant, or pgvector — extra infra, extra latency, extra bill. Vectorize collapses all of that into a Worker binding. For Cloudflare-native AI apps, it's the lowest-friction option. If you'd rather not build the pipeline at all, Cloudflare's managed RAG service, [[Cloudflare AI Search|AI Search]] (formerly AutoRAG; its binding replaced the old `env.AI.autorag()` API), went GA on October 1, 2026. It handles ingestion, storage, embeddings, reranking and hybrid search (vector plus full-text, on by default) for you; usage billing starts November 1, 2026. Vectorize stays the lower-level option when you want control over the index. ## Common Use Cases - **Semantic search** over documents, products, code - **RAG** for chatbots and assistants (embed → retrieve → prompt) - **Recommendations** based on embedding similarity - **Deduplication** and clustering of unstructured content - **Personalization** — user-preference vectors matched against item vectors ## Limits Worth Knowing - Max vector dimensions: 1536 (covers OpenAI ada-002, BGE, most common models) - Max vectors per index: 20M since August 4, 2026 (10M in January, 5M before that) - topK: up to 50 results with values or metadata (since March 2026), 100 without - Metadata filtering supported at query time (up to 10 metadata indexes per index) - Free tier: 30M queried and 5M stored vector dimensions per month - Paid: 50M queried dimensions included, then $0.01 per million; 10M stored dimensions included, then $0.05 per 100 million ## References - Vectorize home: https://developers.cloudflare.com/vectorize/ - Workers AI tutorials (RAG included): https://developers.cloudflare.com/workers-ai/guides/tutorials/ - Limits: https://developers.cloudflare.com/vectorize/platform/limits/ - Pricing: https://developers.cloudflare.com/vectorize/platform/pricing/ - Changelog, topK raised to 50 (2026-03-16): https://developers.cloudflare.com/changelog/post/2026-03-16-topk-limit-increased-to-50/ - Changelog, 20 million vectors per index (2026-08-04): https://developers.cloudflare.com/changelog/post/2026-08-04-index-capacity-20-million/ - Changelog, AI Search namespace bindings replace `env.AI.autorag()` (2026-04-16): https://developers.cloudflare.com/changelog/post/2026-04-16-ai-search-namespace-binding/ - Changelog, AI Search GA (2026-10-01): https://developers.cloudflare.com/changelog/post/2026-10-01-ai-search-generally-available/ ## Related - [[Cloudflare]] - [[Cloudflare Workers]] - [[Cloudflare Workers AI]] - [[Cloudflare AI Search]] - [[Cloudflare D1]] - [[Cloudflare R2]] - [[Cloudflare Agents SDK]] - [[Retrieval-Augmented Generation (RAG)]] - [[Wrangler]]