chore(web)(deps): bump the web-dependencies group across 1 directory with 16 updates

[dependabot]

Bumps the web-dependencies group with 16 updates in the /web directory:

Package	From	To
@tanstack/ai	0.21.3	0.46.0
@tanstack/ai-anthropic	0.10.3	0.16.7
@tanstack/ai-client	0.11.8	0.24.0
@tanstack/ai-gemini	0.10.10	0.24.1
@tanstack/ai-openai	0.10.5	0.20.0
@tanstack/ai-react	0.11.8	0.20.0
@tanstack/devtools-event-client	0.4.4	0.5.0
@tanstack/devtools-vite	0.7.2	0.8.4
@tanstack/match-sorter-utils	8.19.4	9.1.2
@tanstack/react-query	5.100.14	5.101.4
@tanstack/react-router	1.170.8	1.170.31
@tanstack/react-table	8.21.3	9.1.2
docx-preview	0.3.7	0.4.0
katex	0.17.0	0.18.4
@testing-library/jest-dom	6.10.0	7.0.1
@types/qrcode	1.5.5	1.5.6

Updates @tanstack/ai from 0.21.3 to 0.46.0

Release notes

Sourced from @​tanstack/ai's releases.

@​tanstack/ai-client@​0.31.0

Minor Changes

• #1309 21775ee - Add @tanstack/ai-cloudflare: a Cloudflare adapter for Workers AI chat, summarization, embeddings, image generation, text-to-speech, and transcription over the env.AI binding or the REST API, with AI Gateway routing (gateway option and cloudflareGateway() helper for other providers). @tanstack/ai learns the cloudflare max-tokens key for summarize, lets defineByokProvider declare companion credentials with with, and adds getByokKeys(request, { name: provider }) to @tanstack/ai/byok/server. @tanstack/ai-client's defineByok takes providers: a send for a provider with companions (Cloudflare token plus account id) carries every x-byok-* header and prompts for each missing value.

• #1302 82ced0f - Add the registerWebMCPTools registrar to @tanstack/ai-client. Each framework package adds a lifecycle wrapper through useWebMCPTools, createWebMCPTools, or injectWebMCPTools.

Patch Changes

• Updated dependencies [21775ee]:
  • @​tanstack/ai@​0.53.0

@​tanstack/ai-client@​0.30.0

Minor Changes

• #1252 a4ab03e - Add typed headless createChatUI() adapters on @tanstack/ai-react/ui, @tanstack/ai-solid/ui, @tanstack/ai-vue/ui, and @tanstack/ai-svelte/ui. Chat options control the types of message parts, tools, structured output, and interrupts. toolsComponents requires a component for every tool name, and interruptsComponents.generic one for every registered interrupt id. InterruptProps pins a tool approval or a registered generic interrupt. Old Chat orchestration stays importable and deprecated until 1.0.

@​tanstack/ai-gemini@​0.29.1

Patch Changes

• Updated dependencies [21775ee]:
  • @​tanstack/ai@​0.53.0

@​tanstack/ai-gemini@​0.29.0

Minor Changes

• #1312 5305d32 - Accept baseURL and defaultHeaders on every adapter's client config so one gateway config (Cloudflare AI Gateway, Vercel AI Gateway, a corporate proxy) can be spread into any adapter. The vendor-specific names (httpOptions, serverURL, host, baseUrl, headers) keep working. Bedrock's Converse adapter now applies defaultHeaders too.

Patch Changes

• Updated dependencies [819e77c]:
  • @​tanstack/ai@​0.52.3

@​tanstack/ai-gemini@​0.28.0

Minor Changes

• #1301 307b7ec - Add Gemini 3.8 Flash (gemini-3.8-flash) with multimodal input, thinking, structured output, caching, built-in tools, and agentic video understanding.

  Limit its thinking levels to low, medium, and high in both the standard and Interactions adapters.

Patch Changes

• Updated dependencies [452d6a4]:
  • @​tanstack/ai@​0.52.2

@​tanstack/ai-gemini@​0.27.0

Minor Changes

• #1294 4964221 - Add agentic video understanding to the Gemini text adapter.

... (truncated)

Changelog

Sourced from @​tanstack/ai's changelog.

0.46.0

Minor Changes

• #896 41a5d18 - Add a self-describing billed field to TokenUsage so billed quantities carry the unit they are counted in (#816). usage.billed is { quantity, unit } with a BillingUnit union ('seconds', 'units', 'images', 'tokens', ... open-ended). The deprecated unitsBilled / durationSeconds counts are still populated for backward compatibility. The fal adapters report { quantity, unit: 'units' }, Grok video { quantity, unit: 'seconds' }, the OpenAI/Grok/BytePlus duration-billed transcription paths { quantity, unit: 'seconds' }, BytePlus Seedream images { quantity, unit: 'images' }, BytePlus Seedance video { quantity, unit: 'tokens' }, and Cohere/OpenRouter rerank { quantity, unit: 'units' } (search units). Persistence sums billed when both reports use the same unit. otelMiddleware emits the pair as tanstack.ai.usage.billed_quantity / tanstack.ai.usage.billed_unit span attributes.

• #969 ecd12a4 - WebSocket transport: a full-duplex, resumable third transport alongside SSE and NDJSON, reusing the same delivery-durability seam.

  On the server, @tanstack/ai adds toWebSocketStream(socket, request, { onRun, durability, batch, heartbeatMs, idleTimeoutMs, debug }) — a portable core that pumps a conversation over an already-accepted WHATWG WebSocketLike server socket (Node via ws, Bun, etc.), and toWebSocketResponse(request, { onRun, … }), a thin wrapper that upgrades via WebSocketPair and returns a 101 Response on Cloudflare Workers/Durable Objects (it throws elsewhere, pointing you at toWebSocketStream). Because one socket outlives many chat() turns (client-tool resubmits, follow-up user messages), you pass an onRun(ctx) => AsyncIterable<StreamChunk> factory instead of a prebuilt stream — the helper calls it per inbound RunAgentInput frame. The socket is conversation-scoped: it stays open across turns and closes on client close or the idle timeout (which never fires while a turn is still streaming), with a periodic { type: 'ping' } heartbeat. An { type: 'abort', runId } control frame aborts only that turn, leaving the socket open. A turn that fails is surfaced to the client as a live RUN_ERROR frame, mirroring the HTTP transports. Durability is keyed per turn and reuses the existing durableStreamSource, so server→client frames carry the same { id, chunk } envelope as NDJSON. resumeWebSocketStream(socket, { adapter }) and resumeWebSocketResponse({ adapter }) replay a run read-only from the durability log (no model call).

  On the client, webSocket(url, options) (in @tanstack/ai-client, re-exported from @tanstack/ai-react, -solid, -vue, -svelte, and -angular) is a full-duplex subscribe + send connection adapter for useChat. send() writes a RunAgentInput frame; subscribe() yields inbound chunks, ignores heartbeats, unwraps durable envelopes, and auto-reconnects a dropped durable run by reopening with ?runId=&offset= (browsers can't set a Last-Event-ID handshake header, so the offset rides in the URL). The reconnect bookkeeping (offset de-dupe, no-progress ceiling → StreamReconnectLimitError) is shared with the HTTP adapters via the new createReconnectTracker, and a fatal drop surfaces to the consumer (StreamReadError / StreamReconnectLimitError) instead of hanging. Aborting a run (stop() in useChat) sends the { type: 'abort', runId } frame so the server cancels the turn instead of generating to completion, and joinRun() opens its own replay socket so a rejoin never collides with the live conversation socket.

Patch Changes

• #1147 4599019 - Preserve reasoning in server chat message history and interrupt snapshots.

• #932 3eda66c - Classify Anthropic, Gemini, and OpenAI native tools with stable runtime discriminators so ordinary functions can use the same public names without selecting provider-native behavior. Native tools must come from the adapter factory (webSearchTool(), googleSearchTool(), and the rest). A reserved name alone does not select a native converter. chat() throws DuplicateToolNameError when a factory tool and a custom function share the same public name.

... (truncated)

Commits

• 25a9908 ci: Version Packages (#1152)
• 3eda66c fix: distinguish custom and native provider tools (#932)
• 4599019 fix(ai): preserve server chat reasoning (#1147)
• ecd12a4 feat: WebSocket transport — full-duplex resumable streams (#969)
• 41a5d18 feat: add self-describing billed usage to TokenUsage (#896)
• 2822885 ci: Version Packages (#1142)
• 5701bf5 docs: regenerate API documentation (#1141)
• b97a11b fix(ai | ai-groq): recover provider-rejected tool calls (#1139)
• b2b1c43 ci: Version Packages (#1118)
• a8454a7 fix(ai-gemini): add GA image model ids and type sizes per model (#1104)
• Additional commits viewable in compare view

Updates @tanstack/ai-anthropic from 0.10.3 to 0.16.7

Changelog

Sourced from @​tanstack/ai-anthropic's changelog.

0.16.7

Patch Changes

• #932 3eda66c - Classify Anthropic, Gemini, and OpenAI native tools with stable runtime discriminators so ordinary functions can use the same public names without selecting provider-native behavior. Native tools must come from the adapter factory (webSearchTool(), googleSearchTool(), and the rest). A reserved name alone does not select a native converter. chat() throws DuplicateToolNameError when a factory tool and a custom function share the same public name.

  Previously the converters picked provider-native behavior by tool.name. Tool names are public application identifiers, so a plain function called web_search, google_search, or code_execution was routed into a native converter: it lost its inputSchema and was sent as a provider-only payload (and on Anthropic could also flip on code_execution / skills beta headers). Native tools are now identified by adapter-owned metadata, which converters strip before building the wire payload, so provider API versions stay confined to the wire converters.

  Also preserves Anthropic webSearchTool options (max_uses, allowed_domains, blocked_domains, user_location, cache_control) on the wire payload.

  Also fixes googleSearchTool({ searchTypes: … }) being silently dropped on the experimental geminiTextInteractions() adapter. The Interactions converter read a snake_case search_types array, but the public factory takes the Generate Content shape (GoogleSearch.searchTypes: { webSearch?, imageSearch? }), so the field never matched and every request fell back to the provider default of web-search-only. The camelCase config is now translated to the Interactions wire list.

• Updated dependencies [41a5d18, 4599019, 3eda66c, ecd12a4]:

  • @​tanstack/ai@​0.46.0

0.16.6

Patch Changes

• #1136 b7928f2 - Register claude-opus-5 and claude-opus-5-fast on the per-model tool-capabilities map so they match ANTHROPIC_MODELS.

• #1048 bc8c5e8 - Update model metadata from OpenRouter API

• Updated dependencies [d10dfe6, eda82cc, c63319e, b09e010, 0fb8263]:

  • @​tanstack/ai@​0.45.0

0.16.5

Patch Changes

• #1062 ad62202 - Forward top-level cache_control from modelOptions in the Anthropic text adapter.

  Anthropic's Messages API accepts cache_control as a request-level parameter (it auto-places the cache breakpoint on the last cacheable block), but the adapter's modelOptions allowlist omitted it, so any value passed was silently stripped and logged as a dropped key. Per-block caching via systemPrompts[].metadata.cache_control is unchanged.

• #1071 ea9c077 - fix: publish internal dependency ranges as ^x.y.z instead of exact pins

  Internal dependencies on other TanStack AI packages used workspace:* in dependencies and peerDependencies. pnpm rewrites that to an exact version at publish time, so a released package asked for e.g. @tanstack/ai-utils@0.4.0 rather than ^0.4.0.

  Two consequences for consumers:

  • Duplicate copies. An exact pin cannot dedupe. Installing a newer @tanstack/ai alongside a package pinned to the previous patch produced two copies in the tree, which breaks instanceof checks and module-level state, and inflates bundles.
  • Unsatisfiable peers. An exactly pinned peerDependency conflicts the moment the internal package ships its next patch, forcing consumers into overrides or --legacy-peer-deps.

... (truncated)

Commits

• 25a9908 ci: Version Packages (#1152)
• 3eda66c fix: distinguish custom and native provider tools (#932)
• b2b1c43 ci: Version Packages (#1118)
• b7928f2 feat(ai-grok): catch up Grok Imagine image, video, and voice modes (#1136)
• bc8c5e8 chore: sync model metadata from OpenRouter (#1048)
• d45f976 chore(deps): update build & test tooling (#1124)
• e02b661 ci: Version Packages (#1059)
• ea9c077 fix: publish internal dependency ranges as ^x.y.z instead of exact pins (#1071)
• ad62202 fix(ai-anthropic): forward top-level cache_control from modelOptions (#1062)
• 7f3c5fe ci: Version Packages (#968)
• Additional commits viewable in compare view

Updates @tanstack/ai-client from 0.11.8 to 0.24.0

Changelog

Sourced from @​tanstack/ai-client's changelog.

0.24.0

Minor Changes

• #969 ecd12a4 - WebSocket transport: a full-duplex, resumable third transport alongside SSE and NDJSON, reusing the same delivery-durability seam.

  On the server, @tanstack/ai adds toWebSocketStream(socket, request, { onRun, durability, batch, heartbeatMs, idleTimeoutMs, debug }) — a portable core that pumps a conversation over an already-accepted WHATWG WebSocketLike server socket (Node via ws, Bun, etc.), and toWebSocketResponse(request, { onRun, … }), a thin wrapper that upgrades via WebSocketPair and returns a 101 Response on Cloudflare Workers/Durable Objects (it throws elsewhere, pointing you at toWebSocketStream). Because one socket outlives many chat() turns (client-tool resubmits, follow-up user messages), you pass an onRun(ctx) => AsyncIterable<StreamChunk> factory instead of a prebuilt stream — the helper calls it per inbound RunAgentInput frame. The socket is conversation-scoped: it stays open across turns and closes on client close or the idle timeout (which never fires while a turn is still streaming), with a periodic { type: 'ping' } heartbeat. An { type: 'abort', runId } control frame aborts only that turn, leaving the socket open. A turn that fails is surfaced to the client as a live RUN_ERROR frame, mirroring the HTTP transports. Durability is keyed per turn and reuses the existing durableStreamSource, so server→client frames carry the same { id, chunk } envelope as NDJSON. resumeWebSocketStream(socket, { adapter }) and resumeWebSocketResponse({ adapter }) replay a run read-only from the durability log (no model call).

  On the client, webSocket(url, options) (in @tanstack/ai-client, re-exported from @tanstack/ai-react, -solid, -vue, -svelte, and -angular) is a full-duplex subscribe + send connection adapter for useChat. send() writes a RunAgentInput frame; subscribe() yields inbound chunks, ignores heartbeats, unwraps durable envelopes, and auto-reconnects a dropped durable run by reopening with ?runId=&offset= (browsers can't set a Last-Event-ID handshake header, so the offset rides in the URL). The reconnect bookkeeping (offset de-dupe, no-progress ceiling → StreamReconnectLimitError) is shared with the HTTP adapters via the new createReconnectTracker, and a fatal drop surfaces to the consumer (StreamReadError / StreamReconnectLimitError) instead of hanging. Aborting a run (stop() in useChat) sends the { type: 'abort', runId } frame so the server cancels the turn instead of generating to completion, and joinRun() opens its own replay socket so a rejoin never collides with the live conversation socket.

Patch Changes

• #1154 f6b9d89 - Mint omitted threadId after the view mounts, not during render. DevTools binds the hook row to threadId. Persistence that is on (true or a storage adapter) requires a threadId at compile time, and throws at runtime if it is missing. Chat and generation clients no longer accept a separate id option. Use threadId.

• Updated dependencies [41a5d18, 4599019, 3eda66c, ecd12a4]:

  • @​tanstack/ai-event-client@​0.9.0
  • @​tanstack/ai@​0.46.0

... (truncated)

Commits

• 25a9908 ci: Version Packages (#1152)
• f6b9d89 fix(ai-client): use threadId as the only client identity (#1154)
• ecd12a4 feat: WebSocket transport — full-duplex resumable streams (#969)
• 5701bf5 docs: regenerate API documentation (#1141)
• b2b1c43 ci: Version Packages (#1118)
• f492bc9 fix(ai-client): preserve client-tool continuation ownership (#1122)
• d45f976 chore(deps): update build & test tooling (#1124)
• e02b661 ci: Version Packages (#1059)
• ea9c077 fix: publish internal dependency ranges as ^x.y.z instead of exact pins (#1071)
• a8bf8bc fix(ai-client): correlate synthesized resume terminals to request run (#1053)
• Additional commits viewable in compare view

Updates @tanstack/ai-gemini from 0.10.10 to 0.24.1

Changelog

Sourced from @​tanstack/ai-gemini's changelog.

0.24.1

Patch Changes

• #932 3eda66c - Classify Anthropic, Gemini, and OpenAI native tools with stable runtime discriminators so ordinary functions can use the same public names without selecting provider-native behavior. Native tools must come from the adapter factory (webSearchTool(), googleSearchTool(), and the rest). A reserved name alone does not select a native converter. chat() throws DuplicateToolNameError when a factory tool and a custom function share the same public name.

  Previously the converters picked provider-native behavior by tool.name. Tool names are public application identifiers, so a plain function called web_search, google_search, or code_execution was routed into a native converter: it lost its inputSchema and was sent as a provider-only payload (and on Anthropic could also flip on code_execution / skills beta headers). Native tools are now identified by adapter-owned metadata, which converters strip before building the wire payload, so provider API versions stay confined to the wire converters.

  Also preserves Anthropic webSearchTool options (max_uses, allowed_domains, blocked_domains, user_location, cache_control) on the wire payload.

  Also fixes googleSearchTool({ searchTypes: … }) being silently dropped on the experimental geminiTextInteractions() adapter. The Interactions converter read a snake_case search_types array, but the public factory takes the Generate Content shape (GoogleSearch.searchTypes: { webSearch?, imageSearch? }), so the field never matched and every request fell back to the provider default of web-search-only. The camelCase config is now translated to the Interactions wire list.

• Updated dependencies [41a5d18, 4599019, 3eda66c, ecd12a4]:

  • @​tanstack/ai@​0.46.0

0.24.0

Minor Changes

• #1104 a8454a7 - Add the GA Gemini native image model ids and give each native image model its own size type.

  gemini-3.1-flash-image-preview and gemini-3-pro-image-preview were shut down on 2026-06-25 and now 404. Their GA replacements — gemini-3.1-flash-image and gemini-3-pro-image — are now the primary ids. The -preview ids remain in the model union as aliases so existing code keeps compiling; gemini-2.5-flash-image stays fully supported ahead of its 2026-10-02 shutdown.

  Sizes were a single flat union ({8 ratios}_{1K|2K|4K}) applied to every native model. Google documents four different sets, so each model now maps to its own:

  model	aspect ratios	resolutions
  gemini-3.1-flash-image (+ -preview)	14	512 1K 2K 4K
  gemini-3.1-flash-lite-image	14	1K
  gemini-3-pro-image (+ -preview)	10	1K 2K 4K
  gemini-2.5-flash-image	10	none — bare ratio, e.g. '16:9'

  4:5 and 5:4 are now accepted on every native model (Google lists them for all four; the old union omitted them). 9:21 is deliberately still rejected — it exists on Vertex/Cloud only and the Gemini API rejects it.

  Runtime behaviour changes in two places. The rest of the change is types-only, but these two are real wire-format deltas:

  • parseNativeImageSize() now accepts a bare aspect ratio. Previously '16:9' failed to parse, so the adapter omitted imageConfig entirely and the model picked its own aspect ratio; it now parses to { aspectRatio: '16:9' } and the adapter sends imageConfig.aspectRatio = '16:9'. A JavaScript caller — or a TypeScript caller whose size is computed at runtime and widened to string — that already passed a bare ratio will get a differently-framed image after upgrading, with no compile or runtime error.
  • Migrating a gemini-2.5-flash-image call from '16:9_1K' to the now-required bare '16:9' drops imageSize from the generateContent request. That is intended: Google publishes no image_size value or default for this model, so the adapter no longer guesses a tier the API never documented.

  BREAKING (types only): size combinations the selected model never supported no longer compile. No model id was removed.

  • gemini-3.1-flash-lite-image: 2K and 4K are rejected (the model only emits 1K). Use '<ratio>_1K'.
  • gemini-3-pro-image / gemini-3-pro-image-preview: the extreme banner ratios 1:4 4:1 1:8 8:1 are rejected (Gemini 3.1 Flash Image only), as is the 512 tier.
  • gemini-2.5-flash-image: any _1K / _2K / _4K suffix is rejected — pass the bare ratio ('16:9', not '16:9_1K') — as are the four extreme banner ratios.
  • GeminiNativeImageSize is now the union of the per-model types rather than one flat template literal. It was not previously reachable from the package entry point, so this is a new export rather than a changed one.

  New type exports, so the per-model narrowing is nameable and not just inferred at the call site: GeminiImageModelSizeByName, GeminiStandardImageAspectRatio, GeminiExtendedImageAspectRatio, Gemini31FlashImageSize, Gemini31FlashLiteImageSize, Gemini3ProImageSize, Gemini25FlashImageSize, GeminiNativeImageSize.

  Two caveats worth knowing before you rely on this.

  • No in-editor deprecation warning on the dead -preview ids. The @deprecated tags live on module-private model-metadata consts, and GeminiImageModels is projected out of a const array ((typeof GEMINI_IMAGE_MODELS)[number]), which collapses to bare string literals — JSDoc does not survive that projection. So geminiImage('gemini-3-pro-image-preview') still compiles cleanly with no strikethrough and no hint, and fails only at request time. Grep your codebase for -image-preview rather than expecting the compiler to flag it.
  • gemini-3.1-flash-lite-image's four extreme ratios (1:4 4:1 1:8 8:1) are partially inferred. Unlike the other three native models, Flash Lite has no per-model ratio table on the Gemini API guide. The 14-value set rests on the Cloud model page's explicit enumeration plus ai.google.dev's bare "a discrete set of 14 aspect ratios" assertion; the only Gemini-API enumeration for this model is a 10-item bullet prefixed "New aspect ratios", read here as a what's-new list rather than an exhaustive set. If the API rejects those four in practice, this type over-accepts and should narrow to the 10-ratio set.

... (truncated)

Commits

• 25a9908 ci: Version Packages (#1152)
• 3eda66c fix: distinguish custom and native provider tools (#932)
• 5701bf5 docs: regenerate API documentation (#1141)
• b2b1c43 ci: Version Packages (#1118)
• 4f02789 fix(ai-gemini): stop dropping modelOptions on the native image path (#1103)
• a8454a7 fix(ai-gemini): add GA image model ids and type sizes per model (#1104)
• d45f976 chore(deps): update build & test tooling (#1124)
• 9a4124d ci: Version Packages (#1100)
• 03f0f5d feat(ai-gemini): add gemini-3.7-flash (#1105)
• e02b661 ci: Version Packages (#1059)
• Additional commits viewable in compare view

Updates @tanstack/ai-openai from 0.10.5 to 0.20.0

Changelog

Sourced from @​tanstack/ai-openai's changelog.

0.20.0

Minor Changes

• #896 41a5d18 - Add a self-describing billed field to TokenUsage so billed quantities carry the unit they are counted in (#816). usage.billed is { quantity, unit } with a BillingUnit union ('seconds', 'units', 'images', 'tokens', ... open-ended). The deprecated unitsBilled / durationSeconds counts are still populated for backward compatibility. The fal adapters report { quantity, unit: 'units' }, Grok video { quantity, unit: 'seconds' }, the OpenAI/Grok/BytePlus duration-billed transcription paths { quantity, unit: 'seconds' }, BytePlus Seedream images { quantity, unit: 'images' }, BytePlus Seedance video { quantity, unit: 'tokens' }, and Cohere/OpenRouter rerank { quantity, unit: 'units' } (search units). Persistence sums billed when both reports use the same unit. otelMiddleware emits the pair as tanstack.ai.usage.billed_quantity / tanstack.ai.usage.billed_unit span attributes.

Patch Changes

• #932 3eda66c - Classify Anthropic, Gemini, and OpenAI native tools with stable runtime discriminators so ordinary functions can use the same public names without selecting provider-native behavior. Native tools must come from the adapter factory (webSearchTool(), googleSearchTool(), and the rest). A reserved name alone does not select a native converter. chat() throws DuplicateToolNameError when a factory tool and a custom function share the same public name.

  Previously the converters picked provider-native behavior by tool.name. Tool names are public application identifiers, so a plain function called web_search, google_search, or code_execution was routed into a native converter: it lost its inputSchema and was sent as a provider-only payload (and on Anthropic could also flip on code_execution / skills beta headers). Native tools are now identified by adapter-owned metadata, which converters strip before building the wire payload, so provider API versions stay confined to the wire converters.

  Also preserves Anthropic webSearchTool options (max_uses, allowed_domains, blocked_domains, user_location, cache_control) on the wire payload.

  Also fixes googleSearchTool({ searchTypes: … }) being silently dropped on the experimental geminiTextInteractions() adapter. The Interactions converter read a snake_case search_types array, but the public factory takes the Generate Content shape (GoogleSearch.searchTypes: { webSearch?, imageSearch? }), so the field never matched and every request fell back to the provider default of web-search-only. The camelCase config is now translated to the Interactions wire list.

• Updated dependencies [41a5d18, 4599019, 3eda66c, ecd12a4]:

  • @​tanstack/ai@​0.46.0
  • @​tanstack/openai-base@​0.9.15

0.19.1

Patch Changes

• #1048 bc8c5e8 - Update model metadata from OpenRouter API

• Updated dependencies [d10dfe6, eda82cc, c63319e, b09e010, 0fb8263]:

  • @​tanstack/ai@​0.45.0
  • @​tanstack/openai-base@​0.9.13

0.19.0

Minor Changes

• #926 ee07854 - Add a multimodal embed() activity. A single primitive covers one input or a batch — input accepts a string, a text part, an image part, or a fused text+image item written as a nested Array<ContentPart> ([textPart, imagePart], the same shape chat messages use), one vector per item, with the accepted item types narrowed per model at compile time. Top-level dimensions requests Matryoshka output sizes where supported. Results carry embeddings: [{ vector, index }] plus usage when the provider reports it, and embed() participates in generation middleware, debug logging, OTel (gen_ai.operation.name: embeddings), and devtools events like every other activity.

  Provider adapters: openaiEmbedding (text-embedding-3-small/large), geminiEmbedding (gemini-embedding-001), mistralEmbedding (mistral-embed, codestral-embed), ollamaEmbedding (nomic-embed-text and any local model), bedrockEmbedding (Titan Text V2, Titan Multimodal G1 with fused text+image, Cohere Embed v3 on Bedrock), and @tanstack/ai-cohere's cohereEmbedding (embed-v4.0, multimodal text+image with required inputType).

Patch Changes

• #1071 ea9c077 - fix: publish internal dependency ranges as ^x.y.z instead of exact pins

  Internal dependencies on other TanStack AI packages used workspace:* in dependencies and peerDependencies. pnpm rewrites that to an exact version at publish time, so a released package asked for e.g. @tanstack/ai-utils@0.4.0 rather than ^0.4.0.

  Two consequences for consumers:

  • Duplicate copies. An exact pin cannot dedupe. Installing a newer @tanstack/ai alongside a package pinned to the previous patch produced two

... (truncated)

Commits

• 25a9908 ci: Version Packages (#1152)
• 3eda66c fix: distinguish custom and native provider tools (#932)
• 41a5d18 feat: add self-describing billed usage to TokenUsage (#896)
• 5701bf5 docs: regenerate API documentation (#1141)
• b2b1c43 ci: Version Packages (#1118)
• bc8c5e8 chore: sync model metadata from OpenRouter (#1048)
• d45f976 chore(deps): update build & test tooling (#1124)
• e02b661 ci: Version Packages (#1059)
• ee07854 feat: add multimodal embed() activity and provider adapters (#926)
• ea9c077 fix: publish internal dependency ranges as ^x.y.z instead of exact pins (#1071)
• Additional commits viewable in compare view

Updates @tanstack/ai-react from 0.11.8 to 0.20.0

Changelog

Sourced from @​tanstack/ai-react's changelog.

0.20.0

Minor Changes

• #969 ecd12a4 - WebSocket transport: a full-duplex, resumable third transport alongside SSE and NDJSON, reusing the same delivery-durability seam.

  On the server, @tanstack/ai adds toWebSocketStream(socket, request, { onRun, durability, batch, heartbeatMs, idleTimeoutMs, debug }) — a portable core that pumps a conversation over an already-accepted WHATWG WebSocketLike server socket (Node via ws, Bun, etc.), and toWebSocketResponse(request, { onRun, … }), a thin wrapper that upgrades via WebSocketPair and returns a 101 Response on Cloudflare Workers/Durable Objects (it throws elsewhere, pointing you at toWebSocketStream). Because one socket outlives many chat() turns (client-tool resubmits, follow-up user messages), you pass an onRun(ctx) => AsyncIterable<StreamChunk> factory instead of a prebuilt stream — the helper calls it per inbound RunAgentInput frame. The socket is conversation-scoped: it stays open across turns and closes on client close or the idle timeout (which never fires while a turn is still streaming), with a periodic { type: 'ping' } heartbeat. An { type: 'abort', runId } control frame aborts only that turn, leaving the socket open. A turn that fails is surfaced to the client as a live RUN_ERROR frame, mirroring the HTTP transports. Durability is keyed per turn and reuses the existing durableStreamSource, so server→client frames carry the same { id, chunk } envelope as NDJSON. resumeWebSocketStream(socket, { adapter }) and resumeWebSocketResponse({ adapter }) replay a run read-only from the durability log (no model call).

  On the client, webSocket(url, options) (in @tanstack/ai-client, re-exported from @tanstack/ai-react, -solid, -vue, -svelte, and -angular) is a full-duplex subscribe + send connection adapter for useChat. send() writes a RunAgentInput frame; subscribe() yields inbound chunks, ignores heartbeats, unwraps durable envelopes, and auto-reconnects a dropped durable run by reopening with ?runId=&offset= (browsers can't set a Last-Event-ID handshake header, so the offset rides in the URL). The reconnect bookkeeping (offset de-dupe, no-progress ceiling → StreamReconnectLimitError) is shared with the HTTP adapters via the new createReconnectTracker, and a fatal drop surfaces to the consumer (StreamReadError / StreamReconnectLimitError) instead of hanging. Aborting a run (stop() in useChat) sends the { type: 'abort', runId } frame so the server cancels the turn instead of generating to completion, and joinRun() opens its own replay socket so a rejoin never collides with the live conversation socket.

Patch Changes

• #1154 f6b9d89 - Mint omitted threadId after the view mounts, not during render. DevTools binds the hook row to threadId. Persistence that is on (true or a storage adapter) requires a threadId at compile time, and throws at runtime if it is missing. Chat and generation clients no longer accept a separate id option. Use threadId.

• Updated dependencies [41a5d18, 4599019, f6b9d89, 3eda66c, ecd12a4]:

  • @​tanstack/ai@​0.46.0
  • @​tanstack/ai-client@​0.24.0

... (truncated)

Commits

• 25a9908 ci: Version Packages (#1152)
• f6b9d89 fix(ai-client): use threadId as the only client identity (#1154)
• ecd12a4 feat: WebSocket transport — full-duplex resumable streams (#969)
• 5701bf5 docs: regenerate API documentation (#1141)
• b2b1c43 ci: Version Packages (#1118)
• d45f976 chore(deps): update build & test tooling (#1124)
• e02b661 ci: Version Packages (#1059)
• ea9c077 fix: publish internal dependency ranges as ^x.y.z instead of exact pins (#1071)
• 059f8bb ci: Version Packages (#1057)
• 7f3c5fe ci: Version Packages (#968)
• Additional commits viewable in compare view

Updates @tanstack/devtools-event-client from 0.4.4 to 0.5.0

Release notes

Sourced from @​tanstack/devtools-event-client's releases.

@​tanstack/devtools-event-client@​0.5.0

Minor Changes

• #471 b1ac893 - The root export of @tanstack/devtools-event-client now resolves to a no-op outside development (process.env.NODE_ENV !== 'development'), so the real EventClient is tree-shaken out of production bundles by default.

  If you want devtools events to keep working in production, import the real client from the new @tanstack/devtools-event-client/production subpath, which always ships the real implementation. The public API is identical between the two imports.

Comm...

Description has been truncated