fix(deps): update dependency ai-sdk-ollama to v3.8.4

[renovate]

This PR contains the following updates:

Package	Type	Update	Change	OpenSSF
ai-sdk-ollama (source)	dependencies	minor	3.3.0 → 3.8.4

---

Release Notes

jagreehal/ai-sdk-ollama (ai-sdk-ollama)

v3.8.4

Compare Source

Patch Changes

• 6f81c5d: Updated ai-sdk to v6.0.177

v3.8.3

Compare Source

Patch Changes

• 6019088: Updated AI SDK to 6.0.154

v3.8.2

Compare Source

Patch Changes

• 92ca938: Fix TypeScript 6.0 build failure caused by deprecated baseUrl option in tsup DTS generation

v3.8.1

Compare Source

Patch Changes

• aeafdd6: - Update generateText to detect and use the stable output option instead of the deprecated experimental_output when enableToolsWithStructuredOutput is enabled, including a two-phase tools + structured output path and an integration test to guard this behavior.

v3.8.0

Compare Source

Minor Changes

• ed617f1: - Fix: Make OllamaProviderOptions, OllamaChatProviderOptions, and OllamaEmbeddingProviderOptions compatible with the AI SDK's providerOptions (Record<string, JSONObject>). They are now defined as type aliases from Zod schemas (z.infer), so passing e.g. providerOptions: { ollama: { structuredOutputs: true } } into streamText, generateText, or ToolLoopAgent type-checks correctly (fixes #​548).
  • New: Export provider option Zod schemas: ollamaProviderOptionsSchema, ollamaChatProviderOptionsSchema, ollamaEmbeddingProviderOptionsSchema, and ollamaRerankingProviderOptionsSchema for validation or parsing.
  • New: Image generation support via ollama.imageModel(modelId) and OllamaImageModel, using the AI SDK's generateImage() and experimental Ollama image models (e.g. x/z-image-turbo, x/flux2-klein). Supports providerOptions.ollama (e.g. steps, negative_prompt).
  • Change: Reranking model now uses parseProviderOptions from @ai-sdk/provider-utils for providerOptions.ollama; OllamaRerankingProviderOptions is now a type inferred from the exported schema.

v3.7.1

Compare Source

Patch Changes

• 70def97: Support media/image content in tool result messages. When a tool returns output.type === 'content' with image-data, image-url, or file-data (image) parts, the provider now sends them in the tool message's images array to Ollama. Fixes #​527.

v3.7.0

Compare Source

Minor Changes

• 229a396: Use jsonrepair for tool-call argument parsing
  • Tool-call JSON repair: parseToolArguments now uses jsonrepair when JSON.parse fails, so malformed tool-argument strings (trailing commas, unquoted keys, single quotes) are repaired instead of returning {}. Same logic is used when converting messages in convertToOllamaChatMessages.
  • Quoted/double-encoded JSON: When the model returns a quoted JSON string (e.g. "{\"query\":\"weather\"}"), the inner JSON is now parsed so tool arguments are not lost.
  • Exports: parseToolArguments and resolveToolCallingOptions are now exported from the main and browser entry points for advanced use.
  • Example: New tool-json-repair-example.ts in examples/node demonstrates tool-argument repair; json-repair-example.ts and README updated to reference it.

v3.6.0

Compare Source

Minor Changes

• 2cd78cc: Structured output / object generation
  • Cascade JSON repair: Object generation repair now uses a cascade—jsonrepair first, then the built-in Ollama-specific repair (enhancedRepairText) for edge cases (Python True/None, URLs with //, smart quotes, etc.). Repair remains on by default; use enableTextRepair: false or a custom repairText to override.
  • Exports: cascadeRepairText and enhancedRepairText are exported for advanced use and custom repair pipelines.
  • Reliability: Type validation after repair so non-JSON or string-wrapped output is rejected when the schema expects an object/array; fallback merge no longer spreads primitives into the result.
  • Docs & examples: READMEs updated with cascade repair description and links; new examples/node/src/test-cascade-repair.ts example for repair behavior (run with --llm for LLM object generation).

v3.5.0

Compare Source

Minor Changes

• 13d0d64: Fix streaming of reasoning (think) content so that fullStream correctly emits reasoning-start, reasoning-delta, and reasoning-end parts with a stable ID when using models with think: true (e.g. DeepSeek-R1, Qwen 3). Ensures the AI SDK can aggregate reasoning content and that multiple reasoning chunks from the API are represented as a single reasoning span. Message conversion now includes reasoning parts when converting assistant messages to Ollama format.

v3.4.0

Compare Source

Minor Changes

• 8ddf295: Updated to ai-sdk to 6.0.70

---

Configuration

📅 Schedule: (in timezone Europe/Berlin)

• 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 is behind base branch, 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.

[renovate]

Renovate Ignore Notification

Because you closed this PR without merging, Renovate will ignore this update (3.8.4). You will get a PR once a newer version is released. To ignore this dependency forever, add it to the ignoreDeps array of your Renovate config.

If you accidentally closed this PR, or if you changed your mind: rename this PR to get a fresh replacement PR.