Release note

Last updated Apr 20, 2026 This release notes section covers regular updates and minor fixes. For major feature releases or significant updates, see the changelog.

2026-10-01

AI Search is generally available

AI Search is generally available. Usage-based billing begins on November 1, 2026, and managed Workers AI embedding and reranking calls are included in AI Search pricing, so they no longer appear on your Workers AI bill or in your AI Gateway logs. Refer to Limits & pricing for rates and included monthly usage. New instances use hybrid search by default. AI Search also supports multimodal embedding models and image queries through the REST API and public endpoint. OCR is available on every account: plain-text or code files and PDFs with OCR enabled can be up to 10 MiB, while PDFs without OCR and other supported formats are limited to 4 MiB. When type is omitted, AI Search infers the instance type from an HTTP or HTTPS URL or an existing R2 bucket. Refer to Data sources.

2026-08-25

Larger custom metadata values

AI Search supports larger custom metadata values within a shared 10 KiB metadata envelope for each vector. The envelope includes system metadata and JSON overhead, and the first 64 UTF-8 bytes of each indexed string remain filterable. Refer to Metadata attributes for details.

2026-08-06

Custom domains, Cloudflare Access, and namespace public endpoints

An AI Search public endpoint can now be served from a hostname that you own, such as https://search.example.com/search. Because a custom domain routes through your own zone, you can protect the endpoint with Cloudflare Access so users authenticate with your identity provider before any request reaches AI Search. A namespace can also expose its own public endpoint, so one URL searches across the instances you choose and merges the results.

2026-08-06

Discover parse type for website crawling

Website data sources support a new discover parse type that starts at the source URL and collects pages from both your sitemaps and the links it finds while crawling, so a site without a complete sitemap can still be indexed.

2026-07-30

Use AI Search with the Agents SDK, Vercel AI SDK, and LangChain

New guides show how to use an AI Search instance from popular frameworks: the Cloudflare Agents SDK, the Vercel AI SDK, and LangChain. The AI SDK integration is a new ai-search-provider package, and the LangChain integration is a new retriever in the existing langchain-cloudflare package. Refer to the Agents section.

2026-07-08

GIF and BMP image support

AI Search now indexes .gif and .bmp image files. They are converted to searchable text with Workers AI Markdown conversion during sync, with no configuration changes. Refer to supported file types for the full list.

2026-07-08

Filter list items by exact object key

The list items endpoint now accepts a key query parameter to return the item matching an exact object key. Keys are unique per data source, so combine key with source to disambiguate when the same key exists across multiple sources.

2026-06-24

Configurable similarity cache duration and on-demand purge

You can now control how long AI Search retains cached responses with the cache_ttl setting, from 10 minutes up to 6 days. The default is now 48 hours, from the previous fixed duration of 30 days. You can also purge all cached responses for an instance on demand. Refer to similarity cache.

2026-06-24

Additional file type support

Built-in storage, R2, and website data sources now support .mdoc, .sql, and .log.gz files. Refer to supported file types for the full list.

2026-06-18

Migration to managed infrastructure complete

All AI Search instances have finished migrating to managed infrastructure. Every instance now includes built-in storage, a built-in vector index, and built-in web crawling, so the current limits apply across all instances. Storage, vector indexing, and Browser Run usage for crawling are included. Managed Workers AI embedding and reranking calls are also included in AI Search pricing. Generation, query rewriting, and external providers continue to use your account and gateway. If your instance previously crawled a website, the dedicated R2 bucket AI Search created in your account is no longer used and can be deleted.

2026-06-10

Manage AI Search namespaces with Wrangler CLI

AI Search now supports namespace-level Wrangler commands to create, list, update, and delete namespaces from the command line. Instance-level commands also accept a --namespace flag to target instances within a specific namespace, and --json output is available for automation and agent workflows. Refer to Wrangler commands for full usage details.

2026-05-12

Automatic migration to managed infrastructure on June 3, 2026

Instances created before April 16, 2026 will be automatically migrated to managed infrastructure on June 3, 2026. The migration may take up to three days with no service downtime expected. No action is required. After migration, instances will include built-in storage, a built-in vector index, and built-in web crawling, and new limits will apply. For full migration details, refer to Limits & pricing.

2026-04-16

Hybrid search

AI Search now supports hybrid search, combining vector and BM25 keyword search in a single query. Configure the tokenizer, keyword match mode, and fusion method per instance. Refer to Search modes for an overview.

2026-04-16

Relevance boosting

Boost search results by metadata fields like timestamp or priority using relevance boosting. Configure up to 3 boost fields per instance or per request.

2026-04-16

Cross-instance search

Search across multiple AI Search instances in a single call using namespace-level search. Results are merged and ranked, with each chunk identifying which instance it came from.

2026-04-16

Built-in storage and vector index

New AI Search instances come with built-in storage and a built-in vector index. Upload files directly to an instance using the Items API or the dashboard without setting up external infrastructure.

2026-04-16

New AI Search Workers bindings

Two new Workers bindings for AI Search. The ai_search_namespaces binding gives access to all instances within a namespace and supports dynamic instance management at runtime. The ai_search binding binds directly to a single instance for simpler use cases.

2026-04-01

Wrangler CLI support for AI Search

Manage AI Search instances from the command line with the wrangler ai-search command namespace. Create, list, update, delete, search, and get usage statistics for instances without leaving your terminal. All commands support --json for structured output that scripts and AI agents can parse directly. Refer to Wrangler commands for full usage details.

2026-03-23

Custom metadata filtering

Define up to 5 custom metadata fields per AI Search instance and filter search results by category, version, or any custom attribute. Attach metadata via R2 custom headers or HTML meta tags.

2026-03-23

Public endpoint, UI snippets, and MCP support

AI Search now supports public endpoints, UI snippets, and MCP, making it easy to add search to your website or connect AI agents.