CherryHQ/cherry-studio · error · Error

Use embeddingModel() for embedding endpoint type

Error message

Use embeddingModel() for embedding endpoint type

What it means

Thrown by createNewApi's createChatModel() when endpointType === 'embedding'. The provider deliberately separates chat-language resolution from embedding resolution: embedding-typed models must go through provider.embeddingModel(), not provider()/languageModel()/chatModel(). The throw converts a silent wrong-SDK instantiation into an immediate, explicit failure.

Source

Thrown at src/main/ai/provider/custom/newapiProvider.ts:130

      provider: `${NEWAPI_PROVIDER_NAME}.chat`,
      url,
      headers: authHeaders,
      fetch: customFetch
    })

  const createChatModel = (modelId: string): LanguageModelV3 => {
    switch (endpointType) {
      case 'anthropic':
        return createAnthropicModel(modelId)
      case 'gemini':
        return createGeminiModel(modelId)
      case 'openai-response':
        return createResponsesModel(modelId)
      case 'openai':
      case 'image-generation':
        return createCompatibleModel(modelId)
      case 'embedding':
        throw new Error('Use embeddingModel() for embedding endpoint type')
      case 'jina-rerank':
        throw new Error('Use rerankingModel() for jina-rerank endpoint type')
      default:
        return createCompatibleModel(modelId)
    }
  }

  const provider = (modelId: string) => createChatModel(modelId)
  provider.specificationVersion = 'v3' as const

  provider.languageModel = createChatModel

  provider.embeddingModel = (modelId: string) =>
    new OpenAICompatibleEmbeddingModel(modelId, {
      provider: `${NEWAPI_PROVIDER_NAME}.embedding`,
      url,
      headers: authHeaders,
      fetch: customFetch

View on GitHub (pinned to 726446b54c)

Solutions

1. Route embedding-typed NewAPI models through provider.embeddingModel(modelId) — not through the chat/language path.
2. Fix the upstream endpoint_type assignment: a chat-capable model should have endpoint_type 'openai'/'anthropic'/'gemini', not 'embedding'.
3. Before resolving a model, check resolveEffectiveEndpoint(provider, model).endpointType and dispatch to the matching provider method.
4. Catch and surface a user-facing message: 'This model is configured as an embedding endpoint; switch the endpoint type or pick a chat model.'

Example fix

// before
const model = newApiProvider.languageModel(modelId) // throws if endpointType==='embedding'
// after — dispatch by endpoint type
const ep = resolveEffectiveEndpoint(provider, model).endpointType
const model = ep === 'embedding'
  ? newApiProvider.embeddingModel(modelId)
  : newApiProvider.languageModel(modelId)

// Dispatch by endpoint type before model resolution
const ep = resolveEffectiveEndpoint(provider, model).endpointType
if (ep === 'embedding') {
  throw new Error('Use provider.embeddingModel() for embedding-typed models')
}

export function isEmbeddingEndpointMisroute(e: unknown): boolean {
  return e instanceof Error && /Use embeddingModel\(\)/.test(e.message)
}

try {
  newApiProvider.languageModel(modelId)
} catch (e) {
  if (isEmbeddingEndpointMisroute(e)) return newApiProvider.embeddingModel(modelId)
  throw e
}

When it happens

Trigger: A NewAPI model record whose endpoint_type is 'embedding' is routed through the chat path — i.e. provider(modelId), provider.languageModel(modelId), or provider.chatModel(modelId) is invoked. Typically caused by the endpoint-type resolver returning 'embedding' for a model the caller intends to use for chat.

Common situations: Misconfigured endpoint_type in the provider/model config (e.g. an embedding model flagged as a chat model), or generic code that always calls provider() without consulting the endpoint type.

Related errors

• Use rerankingModel() for jina-rerank endpoint type
• Image generation job: no async transport for '${sdkConfig.pr
• Provider extension "${id}" not found. Did you forget to regi
• [theme-contract] ${label} contains duplicate names
• [theme-contract] ${label} has an invalid or duplicate surfac

AI-assisted analysis of CherryHQ/cherry-studio@726446b54c (2026-08-12). Data as JSON: /api/errors/1554d4b6885240fb. Report an issue: GitHub.