voyage-ai-provider/CHANGELOG.md

Voyage AI Provider for running Voyage AI models with Vercel AI SDK

# voyage-ai-provider

## 4.0.0

### Major Changes

- 00a3a0f: Rework multimodal/image embeddings to follow the AI SDK convention used by the
  official providers (e.g. Google), and align the provider with the non-deprecated
  `ProviderV3` surface.

  **Breaking changes**
  - Removed the `embed`/`embedMany` typed-input generics. The AI SDK no longer
    supports `embedMany<ImageEmbeddingInput>(...)`; `values` is always `string[]`.
    The exported `TextEmbeddingInput`, `ImageEmbeddingInput`, and
    `MultimodalEmbeddingInput` types have been removed.
  - Multimodal and image content is now passed through
    `providerOptions.voyage.content` — an array aligned to `values` by index, where
    each entry is the non-text parts merged with the text in `values[i]` (use
    `null` for text-only entries, and an empty string value for image-only
    embeddings). This replaces the previous approach of serializing complex objects
    into the `values` array.
  - `imageEmbeddingModel` and `multimodalEmbeddingModel` are now equivalent (Voyage
    exposes a single multimodal endpoint). The image-specific input validation has
    been removed.

  **New / fixed**
  - Added the non-deprecated `embeddingModel` property to the provider and marked
    `textEmbeddingModel` as `@deprecated` (closes #8).
  - Exported `VoyageMultimodalContentPart` for typing `content` entries.
  - Added the latest model IDs: `voyage-4-large`, `voyage-4`, `voyage-4-lite`
    (text) and `voyage-multimodal-3.5` (multimodal), all verified against the live
    API. `voyage-4-nano` is intentionally omitted — it is an open-weight model and
    is not served by the Voyage API.
  - Fixed the reranker model id `rerank-lite-2` → `rerank-2-lite` to match the
    Voyage API/docs.

## 3.0.0

### Major Changes

- 7e645b6: Updated to AI SDK v6
  - Added reranking support

## 3.0.0-beta.0

### Major Changes

- Updated to AI SDK v6
  - Added reranking support

## 2.0.0

### Major Changes

- b7f0922: - update to @ai-sdk/provider and @ai-sdk/provider-utils beta
  - change to zod v4 internally

### Patch Changes

- 9bf57b7: fix beta versioning
- b7f0922: - fix the export
- b7f0922: - export `voyageEmbeddingOptions` and `voyageMultimodalEmbeddingOptions`

## 2.0.0-beta.4

### Patch Changes

- fix beta versioning

## 2.0.0-beta.3

### Patch Changes

- fix the export

## 2.0.0-beta.2

### Patch Changes

- export `voyageEmbeddingOptions` and `voyageMultimodalEmbeddingOptions`

## 2.0.0-beta.1

### Patch Changes

- update to @ai-sdk/provider and @ai-sdk/provider-utils beta
- change to zod v4 internally

## 2.0.0-alpha.0

### Major Changes

- 2.0.0-alpha.0
- Upgrade to @ai-sdk/provider v2
- upgrade to ai V5

## 1.1.0

### Minor Changes

- Add multimodal and image embedding support
  - Added `imageEmbeddingModel()` for image-only embeddings
  - Added `multimodalEmbeddingModel()` for combined text and image embeddings
  - Enhanced type system with `ImageEmbeddingInput`, `MultimodalEmbeddingInput`, and `TextEmbeddingInput`
  - Updated dependencies:
    - @ai-sdk/provider to 1.1.3
    - @ai-sdk/provider-utils to 2.2.8

## 1.0.2

### Patch Changes

- Add new model ids

## 1.0.1

### Patch Changes

- Update version, add voyage-3-large docs.

## 1.0.0

### Major Changes

- 031f440: Remove encodingFormat settings.

## 0.0.2

### Patch Changes

- Update example README.md
- Update version of `@ai-sdk/provider` and `@ai-sdk/provider-utils`
- Added additional settings as `outputDimension` and `outputDtype`.

## 0.0.1

### Patch Changes

- First public release