fix(deps): update dependency @tanstack/ai-ollama to v0.8.13

[renovate]

This PR contains the following updates:

Package	Change	Age	Confidence
@tanstack/ai-ollama (source)	0.6.5 → 0.8.13

---

Release Notes

TanStack/ai (@​tanstack/ai-ollama)

v0.8.13

Compare Source

Patch Changes

• #​862 86fbbdd - Emit a RUN_ERROR event instead of throwing when the Ollama text adapter's chatStream fails before streaming begins (e.g. the initial client.chat call is rejected because the host is unreachable). The error is now surfaced as a structured chunk — with message, code, and any available rawEvent — so failures flow through the stream consistently with other adapters rather than escaping as an unhandled exception.

• Updated dependencies [afba322, e7ad181]:

  • @​tanstack/ai@​0.39.1

v0.8.12

Compare Source

Patch Changes

• Updated dependencies [b628a4d, b628a4d]:
  • @​tanstack/ai@​0.39.0

v0.8.11

Compare Source

Patch Changes

• Updated dependencies [c1a8732]:
  • @​tanstack/ai@​0.38.0

v0.8.10

Compare Source

Patch Changes

• #​844 a6cceba - Republish all packages with their compiled dist/ output.

  Releases 0.33.0–0.36.0 were published without a dist/ directory: the
  release workflow relied on an Nx-cached build whose outputs were not
  materialized to disk before changeset publish packed the tarballs, and
  files: ["dist"] silently includes nothing when dist/ is absent. The
  published packages therefore contained only src/, so every export
  (./dist/esm/*.js) resolved to a missing file and the packages were
  uninstallable.

  The publish step now runs a fresh, cache-bypassing build of all packages
  immediately before publishing, guaranteeing compiled artifacts are present in
  every tarball.

• Updated dependencies [a6cceba]:

  • @​tanstack/ai@​0.37.0
  • @​tanstack/ai-utils@​0.3.1

v0.8.9

Compare Source

Patch Changes

• Updated dependencies [fbd3762]:
  • @​tanstack/ai@​0.36.0

v0.8.8

Patch Changes

• Updated dependencies [c04abd3]:
  • @​tanstack/ai@​0.35.0

v0.8.7

Patch Changes

• #​480 eddfbbd - Bind tool calls to the assistant message in tool-first streams by setting AG-UI's
  parentMessageId on TOOL_CALL_START.

  When a provider streams a tool call before any text, the StreamProcessor had no
  active assistant message to attach it to, so it created one under a temporary local id.
  The later TEXT_MESSAGE_START then carried the real provider message id, forcing a
  mid-stream id change — which destabilizes UIMessage.id and can remount the message
  subtree in useChat (React list keys, etc.). See #​477.

  Every text adapter generates one stable assistant message id per stream and already uses
  it for TEXT_MESSAGE_START; they now also emit it as parentMessageId on
  TOOL_CALL_START. The processor reads chunk.parentMessageId (?? active assistant id)
  so the message is created with the correct id immediately and the subsequent
  TEXT_MESSAGE_START matches — no rename, no remount.

  Fixed across all adapters that emit TOOL_CALL_START (Anthropic, OpenAI Responses +
  Chat Completions via @tanstack/openai-base, OpenRouter, Gemini including the
  experimental text-interactions adapter, and Ollama).

• Updated dependencies [31de22b]:

  • @​tanstack/ai-utils@​0.3.0
  • @​tanstack/ai@​0.34.0

v0.8.6

Patch Changes

• Updated dependencies [2cb0313, 18e5f4d, 21720dd, 243b8fa]:
  • @​tanstack/ai@​0.33.0

v0.8.5

Patch Changes

• Updated dependencies [8fa6cc5, 8fa6cc5]:
  • @​tanstack/ai@​0.32.0

v0.8.4

Patch Changes

• Updated dependencies [07aaf8b]:
  • @​tanstack/ai@​0.31.0

v0.8.2

Patch Changes

• Updated dependencies [ff267a5, 570c08a, 22c9b42, 215b6b4, 7d44569]:
  • @​tanstack/ai@​0.29.0

v0.8.1

Patch Changes

• Updated dependencies [5deda27]:
  • @​tanstack/ai@​0.40.0

v0.8.0

Minor Changes

• #​660 6df32b5 - BREAKING: Sampling options (temperature, topP, maxTokens) have moved off the root of chat() / ai() / generate() and into provider-native modelOptions. There is no longer a generic root-level sampling surface — each provider accepts its own native keys, fully typed per model:

  • OpenAI (Responses): modelOptions: { temperature, top_p, max_output_tokens }
  • Anthropic: modelOptions: { temperature, top_p, max_tokens }
  • Gemini: modelOptions: { temperature, topP, maxOutputTokens }
  • Grok: modelOptions: { temperature, top_p, max_tokens }
  • Groq: modelOptions: { temperature, top_p, max_completion_tokens }
  • Ollama: modelOptions: { options: { temperature, top_p, num_predict } } (nested)
  • OpenRouter (chat): modelOptions: { temperature, topP, maxCompletionTokens }

  Middleware no longer sees temperature/topP/maxTokens as first-class fields on ChatMiddlewareConfig; mutate config.modelOptions (with the provider-native keys above) instead. metadata is unaffected and stays at the root.

  The public OllamaTextProviderOptions type export has also been removed from @tanstack/ai-ollama. modelOptions is now typed per model — use the exported OllamaChatModelOptionsByName map (indexed by model name) or the underlying ChatRequest from the ollama SDK for arbitrary model strings.

  Migrate automatically with the codemod, which resolves the provider from the adapter and rewrites the keys for you:

  pnpm codemod:move-sampling-to-model-options "src/**/*.{ts,tsx}"

  See the Sampling Options migration guide for details.

Patch Changes

• Updated dependencies [6df32b5]:
  • @​tanstack/ai@​0.27.0

v0.7.1

Patch Changes

• Updated dependencies [5d6cd28]:
  • @​tanstack/ai@​0.26.0

v0.7.0

Compare Source

Minor Changes

• #​242 c251038 - Enhanced token usage reporting for every provider.

  TokenUsage is now the single canonical run-usage type. It is defined once in
  @tanstack/ai-event-client (the dependency-free leaf package) and re-exported by
  @tanstack/ai, so the two packages can no longer drift. It carries optional
  detailed breakdowns alongside the core token counts: promptTokensDetails /
  completionTokensDetails (cached, reasoning, audio, and per-modality tokens),
  durationSeconds for duration-billed models (e.g. Whisper transcription),
  providerUsageDetails for provider-specific metrics, and cost / costDetails
  for provider-reported cost — so a single usage shape covers counts, detailed
  breakdowns, and cost.

  TokenUsage is generic over its provider details bag —
  TokenUsage<TProviderDetails = ProviderUsageDetails> — so adapters return a
  strongly-typed providerUsageDetails (e.g. TokenUsage<AnthropicProviderUsageDetails>)
  while generic consumers keep the open-record default. The default,
  ProviderUsageDetails (Record<string, NonNullable<unknown>>), is now exported and
  uses non-nullish values rather than unknown so TokenUsage stays assignable across
  JSON-serialization boundaries (e.g. TanStack Start server-fn return types). Each
  provider's usage
  extractor now returns undefined (rather than fabricating zeroed totals) when
  the provider reports no usage object, so an absent usage is distinguishable
  from a genuine zero-token run.

  @tanstack/ai still exports UsageTotals as a @deprecated alias of
  TokenUsage for backward compatibility; it will be removed in a future release.

  Detailed usage is extracted in one place per SDK surface: OpenAI-compatible
  providers (OpenAI, Grok, Groq) share the extractors in @tanstack/openai-base,
  while Anthropic, Gemini, Ollama, and OpenRouter normalize their own provider
  usage. The devtools surface cached and reasoning token badges per iteration.

  Usage is now unified across every modality, not just text/chat. Image, audio,
  and text-to-speech results report the same canonical TokenUsage (with
  per-modality breakdowns) instead of a minimal inputTokens/outputTokens shape:

  • ImageGenerationResult.usage, AudioGenerationResult.usage, and the new
    TTSResult.usage are now typed as TokenUsage. Breaking: consumers of
    these fields should read promptTokens/completionTokens instead of
    inputTokens/outputTokens. @tanstack/ai-event-client's ImageUsage is now
    a @deprecated alias of TokenUsage.
  • OpenAI/Grok image generation surface the text-vs-image input token breakdown
    (promptTokensDetails), Gemini image/audio/TTS now surface their full
    usageMetadata (previously dropped), and OpenRouter image generation surfaces
    the chat usage it already returns.
  • Bug fixes: Ollama no longer produces NaN totals or discards duration-only
    usage; Anthropic defaults missing output_tokens and no longer emits empty
    promptTokensDetails/providerUsageDetails objects; OpenAI GPT-4o
    transcription reads the real audio/text input token breakdown and never falls
    back to duration billing.

  Cross-adapter usage parity fixes:

  • PromptTokensDetails/CompletionTokensDetails gain a documentTokens field,
    and Gemini now surfaces DOCUMENT modality token counts (e.g. PDF inputs)
    instead of silently dropping them.
  • OpenAI-compatible chat (OpenAI/Grok/Groq via @tanstack/openai-base) now
    surfaces Predicted-Outputs acceptedPredictionTokens/rejectedPredictionTokens
    under providerUsageDetails, matching the OpenRouter adapter (rejected
    prediction tokens are billed).
  • Grok transcription (/v1/stt) now reports durationSeconds, mirroring the
    Whisper-1 path in the OpenAI transcription adapter.

Patch Changes

• Updated dependencies [c251038]:
  • @​tanstack/ai@​0.25.0

v0.6.24

Compare Source

Patch Changes

• Updated dependencies [c1ae8b9, a452ae8, 8036b50]:
  • @​tanstack/ai@​0.24.0

v0.6.23

Compare Source

Patch Changes

• Updated dependencies [980ff9b, d5645cf]:
  • @​tanstack/ai@​0.23.0

v0.6.22

Compare Source

Patch Changes

• Updated dependencies [02f7d04]:
  • @​tanstack/ai@​0.22.0

v0.6.21

Compare Source

Patch Changes

• Refresh package README content and npm metadata for better discoverability. (#​626)

• Updated dependencies [ebeb22e]:

  • @​tanstack/ai@​0.21.2
  • @​tanstack/ai-utils@​0.2.1

v0.6.20

Compare Source

Patch Changes

• Updated dependencies [ec1393d, 188fe11]:
  • @​tanstack/ai@​0.21.0

v0.6.19

Compare Source

Patch Changes

• Tighten TypeScript safety: enable noImplicitOverride, (#​579)
  noFallthroughCasesInSwitch, and useDefineForClassFields in the
  root tsconfig.json; add a typed-ESLint block scoped to
  packages/*/src/** that turns on no-floating-promises,
  no-misused-promises, await-thenable,
  switch-exhaustiveness-check, consistent-type-exports,
  prefer-readonly, and no-non-null-assertion (errors), plus
  no-explicit-any (warning). @ts-ignore and @ts-nocheck are
  disallowed in library source via @typescript-eslint/ban-ts-comment,
  and as unknown as <T> double-casts are blocked by a
  no-restricted-syntax rule (escape hatches available with an inline
  reason). Two flags from the original five-flag set —
  noPropertyAccessFromIndexSignature and exactOptionalPropertyTypes
  — were tried and rolled back: they produced ~500 lines of bracket-
  access and conditional-spread churn without catching any real bugs,
  and exactOptionalPropertyTypes would have forced consumers using
  it themselves to deal with our internals' style preferences.

  User-visible API surface is unchanged; this is a hardening pass to
  keep streaming/agent-loop correctness and discriminated-union
  exhaustiveness honest going forward. See issue #​564.

• Updated dependencies [2ad137b]:

  • @​tanstack/ai@​0.20.1

v0.6.18

Compare Source

Patch Changes

• feat(ai): systemPrompts accept { content, metadata } with adapter-inferred metadata typing (#​575)

  chat({ systemPrompts }) now accepts either a plain string (the existing
  shape — fully backward compatible) or { content, metadata }. The metadata
  field's type is inferred from the adapter via a new
  TSystemPromptMetadata generic on TextAdapter / BaseTextAdapter:

  • @tanstack/ai-anthropic declares AnthropicSystemPromptMetadata →
    users get cache_control autocomplete and type-checking on
    systemPrompts[i].metadata for Anthropic chats.
  • Adapters with no per-prompt metadata (OpenAI, Gemini, Ollama,
    OpenRouter, openai-base) inherit the default never, which means the
    metadata field carries no meaningful value at the call site —
    TypeScript only accepts undefined there. Provider-foreign metadata
    that reaches an adapter via JS / as any is silently dropped, never
    written to the wire.

  import { chat } from '@​tanstack/ai'
  import { anthropicText } from '@​tanstack/ai-anthropic'
  
  // Anthropic — `cache_control` is autocompleted, no `satisfies` needed.
  chat({
    adapter: anthropicText({ apiKey }, 'claude-sonnet-4-6'),
    systemPrompts: [
      {
        content: 'Stable instructions — cache me.',
        metadata: { cache_control: { type: 'ephemeral' } },
      },
      'Volatile per-request instruction.',
    ],
  })
  
  // OpenAI — `metadata` is `never`; only `undefined` is assignable, so the
  // field is effectively unusable. The object form without `metadata` still
  // works for portability.
  chat({
    adapter: openaiText({ apiKey }, 'gpt-4o-mini'),
    systemPrompts: [
      'Plain string.',
      { content: 'Object form without metadata is allowed.' },
    ],
  })

  New exports:

  • @tanstack/ai: SystemPrompt, NormalizedSystemPrompt types and the
    normalizeSystemPrompts() helper adapters use to normalize the wide
    input shape to { content, metadata? } before consumption.
  • @tanstack/ai-anthropic: AnthropicSystemPromptMetadata interface
    (currently exposes cache_control for prompt caching).

  Internal:

  • New TSystemPromptMetadata = never generic on TextAdapter /
    BaseTextAdapter, surfaced via '~types'['systemPromptMetadata']
    for inference at the chat() call site.
  • Anthropic adapter reads metadata.cache_control and attaches it to
    the corresponding TextBlockParam.
  • All other text adapters call normalizeSystemPrompts() and join
    .content for their respective instructions / system /
    systemInstruction fields. Foreign metadata that reaches them via JS
    / as any is dropped (never written to the wire).
  • normalizeSystemPrompts() is the public API boundary and throws
    TypeError (naming the offending index) for object-form entries whose
    content isn't a string — preventing literal "undefined" from
    reaching the model on stale call sites.
  • OpenTelemetry middleware attaches per-prompt metadata as the
    tanstack.ai.system_prompt.metadata JSON span attribute when
    captureContent: true and at least one entry carries metadata, so
    observability backends can distinguish cache hit/miss for Anthropic.
  • @tanstack/ai-event-client mirrors the SystemPrompt shape locally
    (avoids a circular import) and projects metadata away on the devtools
    wire — devtools UI still receives Array<string>.

• Updated dependencies [496db9c]:

  • @​tanstack/ai@​0.20.0

v0.6.17

Compare Source

Patch Changes

• Updated dependencies [2e0e2eb]:
  • @​tanstack/ai@​0.19.0

v0.6.16

Compare Source

Patch Changes

• fix(ai-ollama): inject system prompts into messages array for chat API (#​568)

v0.6.15

Compare Source

Patch Changes

• Updated dependencies [a9d1916, e810153]:
  • @​tanstack/ai@​0.18.0

v0.6.14

Compare Source

Patch Changes

• Streaming structured output across the OpenAI-compatible providers, an OpenAI Chat Completions sibling adapter, a summarize-subsystem unification, and the decoupling of @tanstack/ai-openrouter from the shared OpenAI base. (#​527)

v0.6.13

Compare Source

Patch Changes

• Updated dependencies [87f305c]:
  • @​tanstack/ai@​0.16.0

v0.6.12

Compare Source

Patch Changes

• Internal refactor: every provider now delegates getApiKeyFromEnv / generateId / transformNullsToUndefined / ModelMeta helpers to the new @tanstack/ai-utils package. ai-openai and ai-grok additionally inherit OpenAI-compatible adapter base classes (Chat Completions / Responses text, image, summarize, transcription, TTS, video) from the new @tanstack/openai-base package; ai-groq keeps its own BaseTextAdapter-derived text adapter (Groq uses the groq-sdk, not the OpenAI SDK) but consumes @tanstack/openai-base's schema converter and tool converters. The remaining providers (ai-anthropic, ai-gemini, ai-ollama, ai-openrouter, ai-fal, ai-elevenlabs) only consume @tanstack/ai-utils because they speak provider-native protocols, not OpenAI-compatible ones. No breaking changes — all public APIs remain identical. (#​409)

• Updated dependencies [27c9aeb]:

  • @​tanstack/ai-utils@​0.2.0

v0.6.11

Compare Source

Patch Changes

• Updated dependencies [a4e2c55, 82078bd, b2d3cc1]:
  • @​tanstack/ai@​0.15.0

v0.6.10

Compare Source

Patch Changes

• refactor(ai-ollama): extract tool conversion into src/tools/ matching peer adapters (#​465)

  Tool handling lived inline inside the text adapter with raw type casts. It is now split into a dedicated tool-converter.ts / function-tool.ts pair (mirroring the structure used by ai-openai, ai-anthropic, ai-grok, and ai-groq) and re-exported from the package index as convertFunctionToolToAdapterFormat and convertToolsToProviderFormat. Runtime behavior is unchanged.

• Updated dependencies [54523f5, 54523f5, af9eb7b, 54523f5]:

  • @​tanstack/ai@​0.14.0

v0.6.9

Compare Source

Patch Changes

• Wire each adapter's text, summarize, image, speech, transcription, and video paths through the new InternalLogger from @tanstack/ai/adapter-internals: logger.request(...) before each SDK call, logger.provider(...) for every chunk received, and logger.errors(...) in catch blocks. Migrates all pre-existing ad-hoc console.* calls in adapter catch blocks (including the OpenAI and ElevenLabs realtime adapters) onto the structured logger. No adapter factory or config-shape changes. (#​467)

• Updated dependencies [c1fd96f]:

  • @​tanstack/ai@​0.13.0

v0.6.8

Compare Source

Patch Changes

• Updated dependencies [e32583e]:
  • @​tanstack/ai@​0.12.0

v0.6.7

Compare Source

Patch Changes

• Align stream output with @tanstack/ai's AG-UI-compliant event shapes: emit REASONING_* events alongside STEP_*, thread threadId/runId through RUN_STARTED/RUN_FINISHED, and return flat RunErrorEvent shape. Cast raw events through an internal asChunk helper so they line up with the re-exported @ag-ui/core EventType enum. No changes to adapter factory signatures or config shapes. (#​474)

• Updated dependencies [12d43e5]:

  • @​tanstack/ai@​0.11.0

v0.6.6

Compare Source

Patch Changes

• fix(ai, ai-openai, ai-gemini, ai-ollama): normalize null tool input to empty object (#​430)

  When a model produces a tool_use block with no input, JSON.parse('null') returns null which fails Zod schema validation and silently kills the agent loop. Normalize null/non-object parsed tool input to {} in executeToolCalls, ToolCallManager.completeToolCall, ToolCallManager.executeTools, and the OpenAI/Gemini/Ollama adapter TOOL_CALL_END emissions. The Anthropic adapter already had this fix.

• Updated dependencies [c780bc1]:

  • @​tanstack/ai@​0.10.3

---

Configuration

📅 Schedule: (UTC)

• Branch creation
  • Between 12:00 AM and 03:59 AM, only on Monday (* 0-3 * * 1)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.