build(deps): bump workers-ai-provider from 2.0.0 to 4.0.0

[dependabot]

Bumps workers-ai-provider from 2.0.0 to 4.0.0.

Release notes

Sourced from workers-ai-provider's releases.

workers-ai-provider@4.0.0

Major Changes

• #601 7a6f8dd Thanks @​OskarLebuda! - Add support for AI SDK 7

  Bumps peer dependencies to require ai@^7 and @ai-sdk/provider@^4. This is a breaking change for users on AI SDK 6, who should stay on the previous major version of these packages.

Patch Changes

• #602 2e64811 Thanks @​aninibread! - Docs: point AI Search users to the new dedicated ai-search-provider package for Cloudflare's new ai_search and ai_search_namespaces Workers bindings. createAISearch and createAutoRAG in this package continue to wrap the legacy AutoRAG binding, unchanged.

workers-ai-provider@3.3.1

Patch Changes

• #599 c72bb8b Thanks @​threepointone! - fix: route native unified-billing catalog slugs (e.g. deepseek/deepseek-v4-pro) through the run path

  workers-ai-provider@3.2.x began treating any "<vendor>/<model>" id as a third-party AI Gateway catalog slug. This regressed models like deepseek/deepseek-v4-pro that Cloudflare serves natively on the unified-billing run path (env.AI.run): without providers the provider threw at construction, and with providers: [openai] it routed to the gateway universal chat/completions (BYOK) endpoint and returned auth errors.

  deepseek/* is now on the unified run catalog (defaults to env.AI.run), matching pre-3.2 behavior. Unified-billing eligibility is decided per-model by Cloudflare's catalog, so BYOK-only deepseek models (e.g. deepseek/deepseek-chat) still work by opting into the gateway path per call (transport: "gateway" / byok). The rest of the OpenAI-wire long tail (mistral, perplexity, cerebras, openrouter, fireworks) is not on Cloudflare's unified run catalog — env.AI.run returns model-not-found for those — so they remain BYOK gateway-path providers.

  When no providers are configured, a bare "<vendor>/<model>" id is now routed through the same hardened run path as dynamic/* routes: the gateway defaults to the account's "default" gateway (third-party unified billing needs one), cacheTtl/skipCache/metadata/collectLog are applied to the gateway options, and delegate-only options (byok, transport: "gateway", fallback, resume, resume callbacks) now throw a clear error instead of being silently dropped.

workers-ai-provider@3.3.0

Minor Changes

• #590 fe0d182 Thanks @​threepointone! - The AI Gateway delegate gains cross-vendor server-side fallback (fallback: { mode: "server" }) — multiple vendors in one gateway run, with the winner selected via cf-aig-step.

  The gateway delegate now reaches header parity with the run path: the gateway path forwards cacheKey, eventId, requestTimeoutMs, and retries from the gateway options as cf-aig-* headers, and DelegateCallOptions gains two new universal-endpoint controls — byokAlias (cf-aig-byok-alias, select a stored BYOK key by alias) and zdr (cf-aig-zdr, per-request Zero Data Retention override for Unified Billing, applied on both transports).

  Internally, the provider registry, cf-aig-* header building, resumable-stream engine, and Workers AI SSE helpers are now shared across the Cloudflare AI packages (bundled inline — no new dependency for you to install).

• #593 1c6afd0 Thanks @​threepointone! - Native Workers AI failures are now surfaced as AI SDK APICallErrors so the AI SDK's built-in retry (maxRetries) can engage on transient errors.

  Previously the binding path (env.AI.run) threw plain Errors and the REST path threw a generic Error, so the AI SDK never retried them — most notably the common "out of capacity" failure (internal code 3040, HTTP 429) and other 5xx blips just failed the call outright.

  • Binding path: errors thrown by env.AI.run are normalized into an APICallError across every Workers AI model — chat, embedding, image, speech,

... (truncated)

Commits

• fa82946 Version Packages (#613)
• 7a6f8dd Add support for AI SDK 7 (#601)
• 992e10e Merge pull request #602 from aninibread/update-cloudflare-ai-search
• 293f5cd fix(ai-search-provider): forward all generation options, support tools and mu...
• a6c830e define binding as AI_SEARCH
• 2e64811 feat(ai-search-provider): add AI Search provider for the AI SDK
• 976ab71 Merge pull request #600 from cloudflare/changeset-release/main
• 3055d70 Version Packages
• 484f739 Merge pull request #599 from cloudflare/fix/deepseek-run-catalog-routing
• c72bb8b fix(workers-ai-provider): route native unified-billing catalog slugs through ...
• Additional commits viewable in compare view

Maintainer changes

This version was pushed to npm by GitHub Actions, a new releaser for workers-ai-provider since your current version.

Note
Automatic rebases have been disabled on this pull request as it has been open for over 30 days.

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request