CherryHQ/cherry-studio · error · Error

local-embedding provider only supports text embeddings, not

Error message

local-embedding provider only supports text embeddings, not ${capability}

What it means

Thrown by the local-embedding provider when `languageModel` or `imageModel` is called. This provider is embedding-only by design — it exists solely to serve the knowledge base's local text embeddings via transformers.js. The `unsupported` helper throws immediately to make the capability boundary explicit rather than returning a model that would fail later.

Source

Thrown at src/main/ai/provider/custom/localEmbedding/localEmbeddingProvider.ts:53

  /**
   * Unused in-process (no HTTP), but the shared provider config builder does
   * `config.providerSettings.fetch ??= customFetch` across the settings union,
   * so every member must carry this field. See `provider/config.ts`.
   */
  fetch?: FetchFunction
}

/**
 * Embedding-only `ProviderV3`. Language / image models throw, since this
 * provider exists solely to serve the knowledge base's local text embeddings.
 */
// `_settings` is unused (in-process provider, no config) but its type is how the
// extension registry infers this provider's settings type — keep the parameter.
// oxlint-disable-next-line no-unused-vars
export function createLocalEmbeddingProvider(_settings: LocalEmbeddingProviderSettings = {}): ProviderV3 {
  const embeddingModel = (modelId: string): EmbeddingModelV3 => new LocalEmbeddingModel(modelId)
  const unsupported = (capability: string) => (): never => {
    throw new Error(`local-embedding provider only supports text embeddings, not ${capability}`)
  }

  return {
    specificationVersion: 'v3',
    embeddingModel,
    textEmbeddingModel: embeddingModel,
    languageModel: unsupported('language models') as (modelId: string) => LanguageModelV3,
    imageModel: unsupported('image models') as (modelId: string) => ImageModelV3
  }
}

View on GitHub (pinned to 726446b54c)

Solutions

1. Use the local-embedding provider only for text embeddings (`embeddingModel` / `textEmbeddingModel`).
2. Route chat and image requests to a provider that supports them (OpenAI, DashScope, etc.).
3. Fix the model/provider mapping that misrouted a non-embedding call here.

Example fix

// before
const model = localProvider.languageModel('bge-m3') // throws
// after
const model = localProvider.textEmbeddingModel('bge-m3')

if (providerId === LOCAL_EMBEDDING_PROVIDER_ID && capability !== 'embedding') {
  throw new Error(`local-embedding only supports embeddings — route ${capability} elsewhere`)
}

const supportsCapability = (providerId: string, capability: 'embedding' | 'language' | 'image'): boolean =>
  providerId !== LOCAL_EMBEDDING_PROVIDER_ID || capability === 'embedding'

When it happens

Trigger: Calling `provider.languageModel(id)` or `provider.imageModel(id)` on a provider created by `createLocalEmbeddingProvider`; or a model-routing layer dispatching a chat/image request to the local-embedding provider id.

Common situations: A model was tagged with the local-embedding provider id in config but referenced for chat/image; a routing table fell back to the wrong provider; a test reused the local provider for a non-embedding call.

Related errors

• Unknown chat type: ${type}
• MethodNotFound
• MethodNotFound
• MethodNotFound
• MethodNotFound

AI-assisted analysis of CherryHQ/cherry-studio@726446b54c (2026-08-12). Data as JSON: /api/errors/7a8860efe96aaacf. Report an issue: GitHub.