Add maxTokens support to WorkersAi chat and handle non-string responses

[TimoWilhelm]

• Extract maxTokens from options in chatStream and pass as max_tokens to both streaming and non-streaming run. - Update createWorkersAiBindingFetch to stringify object responses when building assistant messages instead of defaulting to empty string.

[changeset-bot]

🦋 Changeset detected

Latest commit: a375d3f

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@cloudflare/tanstack-ai	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[pkg-pr-new]

Open in StackBlitz

npx https://pkg.pr.new/cloudflare/ai/ai-gateway-provider@459

npx https://pkg.pr.new/cloudflare/ai/@cloudflare/tanstack-ai@459

npx https://pkg.pr.new/cloudflare/ai/workers-ai-provider@459

commit: a375d3f

[threepointone]

Thanks for the PR! I've pushed a few additions on top of your commit:

Formatting fix: Broke the long ternary in create-fetcher.ts (line 379) into a multi-line expression for readability.

Changeset: Added .changeset/workers-ai-max-tokens.md (patch bump for @cloudflare/tanstack-ai).

Tests (6 new, all passing):

In workers-ai-adapter.test.ts:

• should forward maxTokens to the binding as max_tokens — verifies maxTokens: 256 becomes max_tokens: 256 on the binding call
• should not send max_tokens when maxTokens is not set — verifies no spurious max_tokens when omitted

In binding-fetch.test.ts:

• should stringify object responses in non-streaming mode — verifies object response gets JSON.stringify'd
• should default to empty string for non-string non-object responses — verifies numeric responses fall through to ""
• should handle null response field gracefully — verifies null doesn't crash
• should pass max_tokens to binding when provided — verifies the binding fetch layer forwards max_tokens

Note: workers-ai-provider already handles both maxTokens and non-string responses correctly, so no changes needed there.