runTools giving BadRequestError: 400 status code (no body)

[prenx4x]

Confirm this is a Node library issue and not an underlying OpenAI API issue

• This is an issue with the Node library

Describe the bug

I am following https://github.com/openai/openai-node/blob/master/examples/tool-call-helpers.ts example with modifications as per my app. The runner is able to make the call, get the results, but during the final call to get final answer, the runner fails and gives me this error -

{
  failed: true,
  error: BadRequestError: 400 status code (no body)
      at APIError.generate (file:///Users/prenx4x/WebstormProjects/hissab/node_modules/.pnpm/openai@5.1.0_zod@3.25.30/node_modules/openai/src/core/error.ts:72:14)
      at OpenAI.makeStatusError (file:///Users/prenx4x/WebstormProjects/hissab/node_modules/.pnpm/openai@5.1.0_zod@3.25.30/node_modules/openai/src/client.ts:400:28)
      at OpenAI.makeRequest (file:///Users/prenx4x/WebstormProjects/hissab/node_modules/.pnpm/openai@5.1.0_zod@3.25.30/node_modules/openai/src/client.ts:616:24)
      at async _ChatCompletionRunner._createChatCompletion (file:///Users/prenx4x/WebstormProjects/hissab/node_modules/.pnpm/openai@5.1.0_zod@3.25.30/node_modules/openai/src/lib/AbstractChatCompletionRunner.ts:235:28)
      at async _ChatCompletionRunner._runTools (file:///Users/prenx4x/WebstormProjects/hissab/node_modules/.pnpm/openai@5.1.0_zod@3.25.30/node_modules/openai/src/lib/AbstractChatCompletionRunner.ts:318:46) {
    status: 400,
    headers: Headers(11) {
      'accept-ranges' => 'none',
      'alt-svc' => 'h3=":443"; ma=2592000,h3-29=":443"; ma=2592000',
      'content-type' => 'application/json; charset=UTF-8',
      'date' => 'Tue, 03 Jun 2025 17:21:16 GMT',
      'server' => 'scaffolding on HTTPServer2',
      'server-timing' => 'gfet4t7; dur=33',
      'transfer-encoding' => 'chunked',
      'vary' => 'X-Origin, Referer, Origin,Accept-Encoding',
      'x-content-type-options' => 'nosniff',
      'x-frame-options' => 'SAMEORIGIN',
      'x-xss-protection' => '0',
      [immutable]: true
    },
    requestID: null,
    error: undefined,
    code: undefined,
    param: undefined,
    type: undefined
  }

Console logs:

msg {
  role: 'assistant',
  tool_calls: [ { function: [Object], id: '', type: 'function' } ],
  parsed: null,
  content: null
}
functionCall {
  arguments: '{"expressions":["20+30"]}',
  name: 'calculate_with_hissab',
  parsed_arguments: null
}
msg {
  role: 'tool',
  tool_call_id: '',
  content: '[{"expression":"20+30","result":"50"}]'
}
functionCallResult [{"expression":"20+30","result":"50"}]

My function -

export async function generate(
  openai: OpenAI,
  model: Models,
  systemInstructions: string,
  userPrompt: string,
): Promise<any> {
  const messages: ChatCompletionMessageParam[] = [
    { role: "system", content: systemInstructions },
    { role: "user", content: userPrompt },
  ];

  const tools = [hissabExpFunction];
  const runner = openai.chat.completions
    .runTools({
      model,
      stream: false,
      tools,
      messages,
    })
    .on("message", (msg) => console.log("msg", msg))
    .on("functionToolCall", (functionCall) =>
      console.log("functionCall", functionCall),
    )
    .on("functionToolCallResult", (functionCallResult) => 
      console.log("functionCallResult", functionCallResult);
     )
    .on("content", (diff) => process.stdout.write(diff));
    
  console.log("Running...");
  const result = await runner.finalChatCompletion(); // This fails and gives error
  
  console.log("Ran...");
  
  console.dir(runner.messages, { depth: null });
  
  console.log("result", result);
    return {
    // some json
      };
}

My Tool definition -

export const hissabExpFunction: RunnableToolFunction<{
  expressions: string[];
}> = {
  type: "function",
  function: {
    name: "calculate_with_hissab",
    description: "Hissab expressions corresponding to the user prompt",
    parameters: {
      type: "object",
      properties: {
        expressions: {
          type: "array",
          description: "List of Hissab expressions",
          items: {
            type: "string",
          },
        },
      },
    },
    function: calcExp,
    parse: JSON.parse,
  },
};

A few other things I noticed -
tool_call_id is blank in msgs after tool call id. And parsed_arguments is null, although not sure what parsed_argument is.
Is this a bug in the code, or Am I doing something wrong here. Thanks.

To Reproduce

use runTools function with function call as per given code.

Code snippets

export const hissabExpFunction: RunnableToolFunction<{
  expressions: string[];
}> = {
  type: "function",
  function: {
    name: "calculate_with_hissab",
    description: "Hissab expressions corresponding to the user prompt",
    parameters: {
      type: "object",
      properties: {
        expressions: {
          type: "array",
          description: "List of Hissab expressions",
          items: {
            type: "string",
          },
        },
      },
    },
    function: calcExp,
    parse: JSON.parse,
  },
};

export const hissabExpFunction: RunnableToolFunction<{
  expressions: string[];
}> = {
  type: "function",
  function: {
    name: "calculate_with_hissab",
    description: "Hissab expressions corresponding to the user prompt",
    parameters: {
      type: "object",
      properties: {
        expressions: {
          type: "array",
          description: "List of Hissab expressions",
          items: {
            type: "string",
          },
        },
      },
    },
    function: calcExp,
    parse: JSON.parse,
  },
};

OS

macOS

Node version

Node v22

Library version

5.1.0

[Edit]: I am using google gemini

[prenx4x]

Digging a bit into this, I found this forum post - https://discuss.ai.google.dev/t/tool-calling-with-openai-api-not-working/60140/5

The issue does seem to be that the tool_call_id is empty.
Should this line -

openai-node/src/lib/AbstractChatCompletionRunner.ts

Line 338 in b7d5408

const tool_call_id = tool_call.id;

be

const tool_call_id = tool_call.id || tool_call.name

so this field is not empty for non-openai LLMs?

[RobertCraigie]

Thanks for the report, where is the tool call ID for gemini then? This seems like a bug with gemini if their API is not compliant?

[prenx4x]

yes, the proper fix should be on gemini's side to generate a tool_call id. I have reported in google dev forum as well about this.
However as a workaround, we could make a change as suggested above as well.

Also small change from above suggested code - it should be

const tool_call_id = tool_call.id || tool_call.function.name

Currently I have to add the tool_call.function.name manually in my code which works in gemini (Technically the tool_call_id can be anything as long it matched with the tool_call response.