error: [handleAbortError] AI response error; aborting request: 403 status code (no body): DeepL custom endpoint

[vi-xl]

Hello there!

I am trying to integrated DeepL API endpoint into LibreChat so people can just translate their sentences at one place in LibreChat.

For this I added the following in librechat.yaml in custom endpoint section in my local docker environment:

custom:
    - name: 'DeepL'
      apiKey: '${DEEPL_API_KEY}'
      baseURL: "https://api-free.deepl.com/v2/translate" 
      models:
        default: [ "translate" ]
      titleConvo: false
      titleModel: 'translate'
      modelDisplayLabel: 'DeepL Translation'
      addParams:
        text: "{{input}}"  # This is where the user input goes
        source_lang: "EN"  # Default source language
        target_lang: "DE"   # Default target language
        auth_key: "${DEEPL_API_KEY}"  # Include the key in the body
      dropParams: [ "stop", "user", "frequency_penalty", "presence_penalty" ]

But I keep getting error: [handleAbortError] AI response error; aborting request: 403 status code (no body).

Looks like the auth key is not getting properly passed in the params provided. The auth key and base url is correct, as I can get the response correctly through curl and postman.

Can someone help here please? Also let me know if I am not using the right approach. Thanks a lot!

[danny-avila]

Is this set correctly in your env file?

DEEPL_API_KEY=your-key

[vi-xl]

Yes, also I tried explicitly entering the API keys instead of ${DEEPL_API_KEY}.

For example:
apiKey: my_api_key

[danny-avila]

the OpenAI SDK appends /chat/completions to the end of the endpoint, is that a real route for your API?

It would be more helpful to share backend logs for what's going on, in particular, error logs that include stack traces, and debug logs.

If you are using docker, they are saved in the ./logs directory at project root.

Reproduce the issue and then check the latest error and debug logs (files starting with "error" and "debug" in name, i.e. logs/error-2024-09-21.log)

[vi-xl]

Ah yes, you are right. /chat/completions does get appended in the end of the endpoint. No, this is not my real route. Is there way to avoid this?

Error log:

_{"headers":{"access-control-expose-headers":"Server-Timing, X-Trace-ID","date":"Mon, 23 Sep 2024 12:42:54 GMT","server-timing":"l7_lb_tls;dur=50, l7_lb_idle;dur=0, l7_lb_receive;dur=0, l7_lb_total;dur=57","strict-transport-security":"max-age=63072000; includeSubDomains; preload","transfer-encoding":"chunked","x-trace-id":"fbb03a46bbdf4badbab737441fca96c7"},"level":"error","message":"[handleAbortError] AI response error; aborting request: 403 status code (no body)","stack":"Error: 403 status code (no body)\n at APIError.generate (/app/api/node_modules/openai/error.js:51:20)\n at OpenAI.makeStatusError (/app/api/node_modules/openai/core.js:263:33)\n at OpenAI.makeRequest (/app/api/node_modules/openai/core.js:306:30)\n at process.processTicksAndRejections (node:internal/process/task_queues:95:5)\n at async ChatCompletionStream._createChatCompletion (/app/api/node_modules/openai/lib/ChatCompletionStream.js:53:24)\n at async ChatCompletionStream._runChatCompletion (/app/api/node_modules/openai/lib/AbstractChatCompletionRunner.js:314:16)","status":403}_

Debug log:

2024-09-23T12:42:54.318Z debug: [BaseClient] userMessage
{
  messageId: "xxxx",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
  conversationId: "xxxx",
  sender: "User",
  text: "Hello!",
  isCreatedByUser: true,
  tokenCount: 6,
}
2024-09-23T12:42:54.318Z debug: [OpenAIClient] chatCompletion
{
  baseURL: "https://api-free.deepl.com/v2/translate",
    modelOptions.model: "translate",
    modelOptions.user: "xxxx",
    modelOptions.stream: true,
    // 1 message(s)
    modelOptions.messages: [{"role":"user","content":"Hello!"}],
}
2024-09-23T12:42:54.319Z debug: [OpenAIClient] chatCompletion: added params
{
    addParams.text: "Hello",
    addParams.source_lang: "EN",
    addParams.target_lang: "DE",
    modelOptions.model: "translate",
    modelOptions.user: "xxxx",
    modelOptions.stream: true,
    // 1 message(s)
    modelOptions.messages: [{"role":"user","content":"Hello!"}],
    modelOptions.text: "Hello",
    modelOptions.source_lang: "EN",
    modelOptions.target_lang: "DE",
}
2024-09-23T12:42:54.319Z debug: [OpenAIClient] chatCompletion: dropped params
{
  dropParams: ,
    modelOptions.model: "translate",
    modelOptions.user: "xxxx",
    modelOptions.stream: true,
    // 1 message(s)
    modelOptions.messages: [{"role":"user","content":"Hello!"}],
    modelOptions.text: "Hello",
    modelOptions.source_lang: "EN",
    modelOptions.target_lang: "DE",
}
2024-09-23T12:42:54.321Z debug: Making request to https://api-free.deepl.com/v2/translate/chat/completions
2024-09-23T12:42:54.328Z debug: [saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2024-09-23T12:42:54.586Z warn: [OpenAIClient.chatCompletion][stream] API error
2024-09-23T12:42:54.587Z error: [handleAbortError] AI response error; aborting request: 403 status code (no body)
2024-09-23T12:42:54.592Z debug: [AskController] Request closed

[vi-xl]

Hey @danny-avila, any idea how we can avoid the appending of /chat/completions for non OpenAI custom endpoint?
Thanks for your help!