chenfei-wu/TaskMatrix · error

OpenAI API error.

Error message

OpenAI API error.

What it means

Returned by planningLLM.get_workflow when the underlying OpenAI chat completion call fails. OpenAIWrapper catches all exceptions and returns status=False, so this string hides the actual cause (auth, quota, Azure misconfig, network).

Source

Thrown at LowCodeLLM/src/planningLLM.py:63

class planningLLM:
    def __init__(self, temperature) -> None:
        self.prefix = PLANNING_LLM_PREFIX
        self.suffix = PLANNING_LLM_SUFFIX
        self.LLM = OpenAIWrapper(temperature)
        self.messages = [{"role": "system", "content": "You are a helpful assistant."}]

    def get_workflow(self, task_prompt):
        '''
        - input: task_prompt
        - output: workflow (json)
        '''
        messages = self.messages + [{'role': 'user', "content": PLANNING_LLM_PREFIX+'\nThe task is:\n'+task_prompt+PLANNING_LLM_SUFFIX}]
        response, status = self.LLM.run(messages)
        if status:
            return self._txt2json(response)
        else:
            return "OpenAI API error."

    def extend_workflow(self, task_prompt, current_workflow, step):
        messages = self.messages + [{'role': 'user', "content": PLANNING_LLM_PREFIX+'\nThe task is:\n'+task_prompt+PLANNING_LLM_SUFFIX}]
        messages.append({'role': 'user', "content": EXTEND_PREFIX+
                         'The current SOP is:\n'+current_workflow+
                         '\nThe step needs to be extended is:\n'+step+
                         PLANNING_LLM_SUFFIX})
        response, status = self.LLM.run(messages)
        if status:
            return self._txt2json(response)
        else:
            return "OpenAI API error."

    def _txt2json(self, workflow_txt):
        ''' convert the workflow in natural language to json format '''
        workflow = []
        try:
            steps = workflow_txt.split('\n')

View on GitHub (pinned to 4b7664f8d3)

Solutions

1. Add logging of the exception in OpenAIWrapper._post_request_chat to surface the real error.
2. Check OPENAIKEY and Azure variables (USE_AZURE, API_BASE, API_VERSION, MODEL) in the server environment.
3. Confirm the installed openai package still exposes openai.ChatCompletion (0.x API); pin openai<1.0 or migrate to openai.ChatCompletion -> client.chat.completions.create.
4. Retry on transient rate-limit errors.

Example fix

# before
response, status = self.LLM.run(messages)
if status: return self._txt2json(response)
else: return "OpenAI API error."
# after - surface wrapper exception and retry once
response, status = self.LLM.run(messages)
if not status:
    response, status = self.LLM.run(messages)  # simple retry
if status: return self._txt2json(response)
else: raise RuntimeError('OpenAI call failed in get_workflow')

import os
missing = [v for v in ('OPENAIKEY',) if not os.environ.get(v)]
if os.environ.get('USE_AZURE','').lower()=='true':
    missing += [v for v in ('API_BASE','API_VERSION','MODEL') if not os.environ.get(v)]
assert not missing, f'missing env: {missing}'

def is_workflow_json(w) -> bool:
    if not isinstance(w, str) or w == 'OpenAI API error.': return False
    import json; json.loads(w); return True

result = llm.get_workflow(task)
if result == 'OpenAI API error.':
    # fix config or retry with backoff; do not parse it as workflow JSON

When it happens

Trigger: POST /api/get_workflow where openai.ChatCompletion.create throws: missing/invalid OPENAIKEY, USE_AZURE=true with missing/wrong API_BASE/API_VERSION/MODEL, rate limits, or network failure.

Common situations: Server started without exporting OpenAI/Azure env vars; Azure deployment name mismatch; openai library version newer than the 0.x ChatCompletion API the code uses; exhausted quota.

Related errors

• OpenAI API error.
• failed to get_workflow, msg:%s, request data:%s
• failed to extend_workflow, msg:%s, request data:%s
• failed to execute, msg:%s, request data:%s
• internal errors

AI-assisted analysis of chenfei-wu/TaskMatrix@4b7664f8d3 (2026-08-27). Data as JSON: /api/errors/39c234f4b312d90f. Report an issue: GitHub.