VolcanoArk requires adapter.

[guorenn]

Which version of the app are you using?

v3.3.9

Which API Provider are you using?

OpenAI Compatible

Which Model are you using?

Doubao-1.5-pro-256k

What happened?

https://www.volcengine.com/docs/82379/1302008

The DouBao model and DeepSeek v3 model from VolcanoArk cannot be used via OpenAI-compatible connections and require API adapter for VolcanoArk's large model service platform.

Steps to reproduce

Relevant API REQUEST output

Additional context

No response

[wangpfcn]

Also 400 One or more parameters specified in the request are not valid.

[samhvw8]

As the docs you provide, it don't have sdk for nodejs
https://www.volcengine.com/docs/82379/1302008

You can follow this to config via openai compatible provider
https://www.volcengine.com/docs/82379/1330626

[DaydreamCoding]

+1 deepseek v3/r1 in ark

[guorenn]

As the docs you provide, it don't have sdk for nodejs https://www.volcengine.com/docs/82379/1302008

You can follow this to config via openai compatible provider https://www.volcengine.com/docs/82379/1330626

Unable to connect via OpenAI compatible, the OpenAI compatible connection returned a 500 error.

[he426100]

According to the document https://www.volcengine.com/docs/82379/1298454 the content in messages should be a string, while cline's request format is
{"role": "user","content": [{"type": "text","text": "[read_file for 'Logger.php']"}]}, so it is only necessary to convert the content to a string.

[guorenn]

No, the document states: "Array of MessageParam".

[he426100]

No, the document states: "Array of MessageParam".

messages is a MessageParam array, MessageParam.content is a String
content: message content, the text generation model only supports the String type.

Here is one solution:

// 火山引擎 API 的配置
const VOLC_API_URL = "https://ark.cn-beijing.volces.com/api/v3/chat/completions";

export default {
  async fetch(request) {
    // 处理CORS预检请求
    if (request.method === 'OPTIONS') {
      return new Response(null, {
        headers: {
          'Access-Control-Allow-Origin': '*',
          'Access-Control-Allow-Methods': 'POST, OPTIONS',
          'Access-Control-Allow-Headers': '*',
          'Access-Control-Max-Age': '86400',
        }
      });
    }

    try {
      // 克隆原始请求用于读取body
      const clonedReq = request.clone();
      const originalBody = await clonedReq.json();
      
      // 处理messages数组
      if (originalBody.messages && Array.isArray(originalBody.messages)) {
        originalBody.messages = originalBody.messages.map(message => {
          // 转换content字段
          if (message.content && Array.isArray(message.content)) {
            // 拼接所有text内容
            message.content = message.content
              .map(item => item.text)    // 提取text值
              .join('\n');               // 用换行符连接
          }
          return message;
        });
      }

      // 创建代理请求
      const proxyReq = new Request(VOLC_API_URL, {
        method: request.method,
        headers: request.headers,
        body: JSON.stringify(originalBody)
      });

      // 发送请求并获取响应
      const response = await fetch(proxyReq);

      // 统一使用流式管道传输
      const { readable, writable } = new TransformStream();
      response.body.pipeTo(writable);

      // 构建最终响应头
      return new Response(readable, {
        status: response.status,
        headers: response.headers
      });
    } catch (error) {
      return new Response(JSON.stringify({
        error: {
          message: error.message,
          type: 'proxy_error'
        }
      }), {
        status: 500,
        headers: {
          'Content-Type': 'application/json',
          'Access-Control-Allow-Origin': '*'
        }
      });
    }
  }
};

[mrubens]

#886 does this look right?

[mrubens]

What's the best way to identify if an openai compatible config is one of these Volcano Ark models? Are there certain model name patterns or API urls to look for?

[he426100]

Not only volcengine, baidu and tencent also require strings; a configuration should be added to set the request data format.

[cexer]

@guorenn @he426100 I'v made a PR to fix this #891, you can try my branch to see if it works.