linshenkx/prompt-optimizer · error · APIError

Cross-origin connection failed: ${error.message}

Error message

Cross-origin connection failed: ${error.message}

What it means

During model listing the fetch failed with 'Failed to fetch' or 'Connection error', and detectCrossOriginError determined the failure is CORS-related (browser blocked a cross-origin request). The original message is wrapped with the 'Cross-origin connection failed:' prefix.

Source

Thrown at packages/core/src/services/llm/adapters/openai-adapter.ts:175

        if (models.length === 0) {
          throw new APIError('API returned empty model list')
        }

        return models
      }

      throw new APIError('Unexpected API response format')
    } catch (error: any) {
      console.error('[OpenAIAdapter] Failed to fetch models:', error)

      // 连接错误处理（包括跨域检测）
      if (error.message && (error.message.includes('Failed to fetch') ||
          error.message.includes('Connection error'))) {
        const isCrossOriginError = this.detectCrossOriginError(error, baseURL)

        if (isCrossOriginError) {
          throw new APIError(`Cross-origin connection failed: ${error.message}`)
        } else {
          throw new APIError(`Connection failed: ${error.message}`)
        }
      }

      // API返回的错误信息
      if (error.response?.data) {
        throw new APIError(`API error: ${JSON.stringify(error.response.data)}`)
      }

      // 其他错误,保持原始信息
      throw new APIError(error.message || 'Unknown error')
    }
  }

  // ===== 参数定义（用于buildDefaultModel） =====

  /**

View on GitHub (pinned to 3e677b1d9f)

Solutions

1. Enable CORS on the target server (e.g. OLLAMA_ORIGINS=* for Ollama; CORS middleware on FastAPI/nginx)
2. Or proxy requests through your own backend so the browser call is same-origin
3. Verify scheme/port/hostname exactly match the allowed origins
4. For production, never call provider APIs with secret keys directly from the browser

Example fix

# before (ollama blocks browser origin)
# browser console: Cross-origin connection failed: Failed to fetch

# after
Ollama: OLLAMA_ORIGINS=http://localhost:5173 ollama serve
# or route via backend proxy /api/llm -> target

// test CORS before first real call
fetch(endpoint, { method: 'OPTIONS', mode: 'cors' }).catch(() => flagEndpointAsCorsBlocked(endpoint))

function isCorsError(e: unknown): boolean {
  return e instanceof APIError && e.message.startsWith('Cross-origin connection failed')
}

try { models = await adapter.getModelsAsync() }
catch (e) {
  if (isCorsError(e)) { instructUserToEnableCors(); models = [] }
  else throw e
}

When it happens

Trigger: Running the app in a browser and calling an OpenAI-compatible endpoint on another origin that doesn't return Access-Control-Allow-Origin — typical for local gateways (localhost:11434, :8000) accessed from a dev server on another port.

Common situations: Ollama/vLLM/llama.cpp called from a web app without CORS enabled, missing CORS middleware on a custom proxy, gateway allowing only certain origins.

Related errors

• Network error: ${error.message}
• Connection failed: ${error.message}
• formatExecutionErrorMessage(error)
• GENERATION_FAILED
• GENERATION_FAILED

AI-assisted analysis of linshenkx/prompt-optimizer@3e677b1d9f (2026-08-27). Data as JSON: /api/errors/7ef8f29f1acd8c81. Report an issue: GitHub.