AI Search: the search primitive for your agents

Cloudflare has rebranded AutoRAG as AI Search and shipped major new capabilities for building search into AI agents. Key additions include hybrid search combining vector and BM25 keyword retrieval with configurable tokenizers, fusion methods, and optional reranking. New instances now come with built-in storage and vector indexes, eliminating the need to manually wire R2 buckets and Vectorize indexes. A new `ai_search_namespaces` binding lets Workers dynamically create, delete, and search across instances at runtime — enabling per-agent, per-customer, or per-tenant search contexts. Cross-instance search allows querying multiple namespaces in a single call with unified ranked results. Metadata-based relevance boosting (e.g., by timestamp or custom fields) is also supported. The service is in open beta with free usage up to defined limits, with unified pricing planned post-beta.