Issue with CrewAI Agent Output Format Validation

CrewAI Community Support
, , ,
1

Issue with CrewAI Agent Output Format Validation

Hi everyone,

I’m trying to run a CrewAI really simple flow with an agent and a task, but I keep getting validation errors related to the output format, even though the agent correctly returns the expected output ("I worked!"). I’m hoping someone can point me in the right direction.

Code

import asyncio
import json
from pydantic import BaseModel
from crewai import Agent, Task
from crewai.flow.flow import Flow, start
from langchain_openai import ChatOpenAI

# ---- STATE ----
class State(BaseModel):
    steps: int = 0
    code: str = ""
    llm: str = "gpt-3.5-turbo"

# ---- FLOW ----
class DiagramFlow(Flow[State]):

    def update_steps(self):
        self.state.steps += 1

    @start()
    def node_1(self):
        class OutputFormat(BaseModel):
            text: str
            extra_text: str

        llm_instance = ChatOpenAI(
            model_name=self.state.llm,
            temperature=0.7,
        )

        # Create an Agent
        agent_name = Agent(
            role="Friendly assistant",
            goal="Return a message that says 'I worked!'",
            backstory="Testing agent call",
            tools=[],
            llm=llm_instance,
            verbose=False,
        )

        task = Task(
            description="Return a string that says: 'I worked!'",
            expected_output="I worked!", # Attempt to solve the problem 1
            output_pydantic=OutputFormat,# Attempt to solve the problem 2
        )

        result = agent_name.kickoff(task, response_format=OutputFormat)

        print("Raw result:", result.raw) # Kinda works
        print("Parsed output:", result.pydantic.text) # Error
        return result

# ---- ENTRY POINT ----
if __name__ == "__main__":
    state = State()
    flow = DiagramFlow(state=state)
    flow.kickoff()

Error

Failed to parse output into response format: 1 validation error for OutputFormat
Invalid JSON: key must be a string at line 1 column 2
[type=json_invalid, input_value="{'type': 'json_schema', ... 'extra_text': 'Success!'}", input_type=str]

Traceback (most recent call last):
  ...
AttributeError: 'NoneType' object has no attribute 'text'

The agent does return:

{'text': 'I worked!', 'extra_text': 'Success!'}

…but CrewAI fails to parse it into the Pydantic model, throwing a json_invalid error.

And i can’t actually take the result from the result variable, not even using result.raw

What I’ve Tried

  • Passing the Pydantic model via output_pydantic and response_format.
  • Making sure the task description and expected output match the returned values.
  • Printing the raw result to verify the agent is returning correct data.

Question

Has anyone faced this json_invalid error when using output_pydantic in CrewAI? It seems like the agent is returning a Python-style dict (' quotes) instead of proper JSON (" quotes), which prevents parsing. Could be that?

Any suggestions on how to make CrewAI accept the output format correctly?

Thanks in advance!

Also

By trying different solutions i have got this other error message, no idea what i have messed up.

Starting LLM setup...
Starting task execution...
Agent failed to reach a final answer. This is likely a bug - please report it.
An unknown error occurred. Please check the details below.                                                                                                                                    
Error details: 35 validation errors for LiteAgentExecutionStartedEvent                                                                                                                        
messages.str                                                                                                                                                                                  
  Input should be a valid string [type=string_type, input_value=[{'role': 'system', 'cont...trigger_context', None)], input_type=list]                                                        
    For further information visit https://errors.pydantic.dev/2.12/v/string_type                                                                                                              
messages.list[dict[str,str]].1
  Input should be a valid dictionary [type=dict_type, input_value=('used_tools', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].2
  Input should be a valid dictionary [type=dict_type, input_value=('tools_errors', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].3
  Input should be a valid dictionary [type=dict_type, input_value=('delegations', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].4
  Input should be a valid dictionary [type=dict_type, input_value=('i18n', I18N(prompt_file=None)), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].5
  Input should be a valid dictionary [type=dict_type, input_value=('name', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].6
  Input should be a valid dictionary [type=dict_type, input_value=('prompt_context', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].7
  Input should be a valid dictionary [type=dict_type, input_value=('description', "\n      ...that says: 'I worked!'"), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].8
  Input should be a valid dictionary [type=dict_type, input_value=('expected_output', 'I worked!'), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].9
  Input should be a valid dictionary [type=dict_type, input_value=('config', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].10
  Input should be a valid dictionary [type=dict_type, input_value=('callback', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].11
  Input should be a valid dictionary [type=dict_type, input_value=('agent', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].12
  Input should be a valid dictionary [type=dict_type, input_value=('context', NOT_SPECIFIED), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].13
  Input should be a valid dictionary [type=dict_type, input_value=('async_execution', False), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].14
  Input should be a valid dictionary [type=dict_type, input_value=('output_json', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].15
  Input should be a valid dictionary [type=dict_type, input_value=('output_pydantic', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].16
  Input should be a valid dictionary [type=dict_type, input_value=('response_model', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].17
  Input should be a valid dictionary [type=dict_type, input_value=('output_file', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].18
  Input should be a valid dictionary [type=dict_type, input_value=('create_directory', True), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].19
  Input should be a valid dictionary [type=dict_type, input_value=('output', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].20
  Input should be a valid dictionary [type=dict_type, input_value=('tools', []), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].21
  Input should be a valid dictionary [type=dict_type, input_value=('security_config', Secur...ngerprint(metadata={}))), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].22
  Input should be a valid dictionary [type=dict_type, input_value=('id', UUID('38c49b6a-f65...48e-b673-6434cf56e2fc')), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].23
  Input should be a valid dictionary [type=dict_type, input_value=('human_input', False), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].24
  Input should be a valid dictionary [type=dict_type, input_value=('markdown', False), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].25
  Input should be a valid dictionary [type=dict_type, input_value=('converter_cls', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].26
  Input should be a valid dictionary [type=dict_type, input_value=('processed_by_agents', set()), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].27
  Input should be a valid dictionary [type=dict_type, input_value=('guardrail', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].28
  Input should be a valid dictionary [type=dict_type, input_value=('guardrails', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].29
  Input should be a valid dictionary [type=dict_type, input_value=('max_retries', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].30
  Input should be a valid dictionary [type=dict_type, input_value=('guardrail_max_retries', 3), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].31
  Input should be a valid dictionary [type=dict_type, input_value=('retry_count', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].32
  Input should be a valid dictionary [type=dict_type, input_value=('start_time', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].33
  Input should be a valid dictionary [type=dict_type, input_value=('end_time', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].34
  Input should be a valid dictionary [type=dict_type, input_value=('allow_crewai_trigger_context', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type                                                                                                                
Traceback (most recent call last):                                                                                                                                                            
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\chatllm.py", line 62, in <module>                                                                                                        
    flow.kickoff()  # 👈 esto ejecuta el flujo async correctamente                                                                                                                            
    ^^^^^^^^^^^^^^                                                                                                                                                                            
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 816, in kickoff                                                                        
    return asyncio.run(_run_flow())                                                                                                                                                           
           ^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                           
  File "C:\Users\Burbu\AppData\Local\Programs\Python\Python312\Lib\asyncio\runners.py", line 195, in run                                                                                      
    return runner.run(main)
           ^^^^^^^^^^^^^^^^                                                                                                                                                                   
  File "C:\Users\Burbu\AppData\Local\Programs\Python\Python312\Lib\asyncio\runners.py", line 118, in run                                                                                      
    return self._loop.run_until_complete(task)                                                                                                                                                
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                
  File "C:\Users\Burbu\AppData\Local\Programs\Python\Python312\Lib\asyncio\base_events.py", line 691, in run_until_complete                                                                   
    return future.result()                                                                                                                                                                    
           ^^^^^^^^^^^^^^^                                                                                                                                                                    
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 814, in _run_flow                                                                      
    return await self.kickoff_async(inputs)                                                                                                                                                   
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                   
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 897, in kickoff_async                                                                  
    await asyncio.gather(*tasks)                                                                                                                                                              
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 966, in _execute_start_method
    result = await self._execute_method(start_method_name, enhanced_method)                                                                                                                   
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                   
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 1067, in _execute_method                                                               
    raise e                                                                                                                                                                                   
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 1032, in _execute_method                                                               
    else method(*args, **kwargs)                                                                                                                                                              
         ^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                              
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow.py", line 998, in enhanced_method                                                                
    return original_method(*args, **kwargs)                                                                                                                                                   
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\flow\flow_wrappers.py", line 84, in __call__                                                               
    return self._meth(self._instance, *args, **kwargs)                                                                                                                                        
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                        
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\chatllm.py", line 51, in node_1                                                                                                          
    result = agent_name.kickoff(task, response_format=OutputFormat)                                                                                                                           
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                           
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\agent\core.py", line 1238, in kickoff                                                                      
    return lite_agent.kickoff(messages)                                                                                                                                                       
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                       
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\lite_agent.py", line 307, in kickoff                                                                       
    raise e                                                                                                                                                                                   
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\lite_agent.py", line 291, in kickoff                                                                       
    return self._execute_core(agent_info=agent_info)                                                                                                                                          
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                          
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\crewai\lite_agent.py", line 313, in _execute_core
    event=LiteAgentExecutionStartedEvent(                                                                                                                                                     
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                     
  File "C:\Users\Burbu\OneDrive - UPV EHU\TFG\codigo\venv\Lib\site-packages\pydantic\main.py", line 250, in __init__                                                                          
    validated_self = self.__pydantic_validator__.validate_python(data, self_instance=self)                                                                                                    
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                    
pydantic_core._pydantic_core.ValidationError: 35 validation errors for LiteAgentExecutionStartedEvent
messages.str
  Input should be a valid string [type=string_type, input_value=[{'role': 'system', 'cont...trigger_context', None)], input_type=list]
    For further information visit https://errors.pydantic.dev/2.12/v/string_type
messages.list[dict[str,str]].1
  Input should be a valid dictionary [type=dict_type, input_value=('used_tools', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].2
  Input should be a valid dictionary [type=dict_type, input_value=('tools_errors', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].3
  Input should be a valid dictionary [type=dict_type, input_value=('delegations', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].4
  Input should be a valid dictionary [type=dict_type, input_value=('i18n', I18N(prompt_file=None)), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].5
  Input should be a valid dictionary [type=dict_type, input_value=('name', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].6
  Input should be a valid dictionary [type=dict_type, input_value=('prompt_context', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].7
  Input should be a valid dictionary [type=dict_type, input_value=('description', "\n      ...that says: 'I worked!'"), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].8
  Input should be a valid dictionary [type=dict_type, input_value=('expected_output', 'I worked!'), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].9
  Input should be a valid dictionary [type=dict_type, input_value=('config', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].10
  Input should be a valid dictionary [type=dict_type, input_value=('callback', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].11
  Input should be a valid dictionary [type=dict_type, input_value=('agent', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].12
  Input should be a valid dictionary [type=dict_type, input_value=('context', NOT_SPECIFIED), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].13
  Input should be a valid dictionary [type=dict_type, input_value=('async_execution', False), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].14
  Input should be a valid dictionary [type=dict_type, input_value=('output_json', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].15
  Input should be a valid dictionary [type=dict_type, input_value=('output_pydantic', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].16
  Input should be a valid dictionary [type=dict_type, input_value=('response_model', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].17
  Input should be a valid dictionary [type=dict_type, input_value=('output_file', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].18
  Input should be a valid dictionary [type=dict_type, input_value=('create_directory', True), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].19
  Input should be a valid dictionary [type=dict_type, input_value=('output', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].20
  Input should be a valid dictionary [type=dict_type, input_value=('tools', []), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].21
  Input should be a valid dictionary [type=dict_type, input_value=('security_config', Secur...ngerprint(metadata={}))), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].22
  Input should be a valid dictionary [type=dict_type, input_value=('id', UUID('38c49b6a-f65...48e-b673-6434cf56e2fc')), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].23
  Input should be a valid dictionary [type=dict_type, input_value=('human_input', False), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].24
  Input should be a valid dictionary [type=dict_type, input_value=('markdown', False), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].25
  Input should be a valid dictionary [type=dict_type, input_value=('converter_cls', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].26
  Input should be a valid dictionary [type=dict_type, input_value=('processed_by_agents', set()), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].27
  Input should be a valid dictionary [type=dict_type, input_value=('guardrail', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].28
  Input should be a valid dictionary [type=dict_type, input_value=('guardrails', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].29
  Input should be a valid dictionary [type=dict_type, input_value=('max_retries', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].30
  Input should be a valid dictionary [type=dict_type, input_value=('guardrail_max_retries', 3), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].31
  Input should be a valid dictionary [type=dict_type, input_value=('retry_count', 0), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].32
  Input should be a valid dictionary [type=dict_type, input_value=('start_time', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].33
  Input should be a valid dictionary [type=dict_type, input_value=('end_time', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type
messages.list[dict[str,str]].34
  Input should be a valid dictionary [type=dict_type, input_value=('allow_crewai_trigger_context', None), input_type=tuple]
    For further information visit https://errors.pydantic.dev/2.12/v/dict_type                                                                                                                
🌊 Flow: DiagramFlow
ID: 7878c5d2-a5e8-4d7b-a368-7de9f326cdb9
├── ✨ Created
├── ✅ Initialization Complete
└── ❌ Failed: node_1
2

Hi.. I think we are missing the agent in the task try

task = Task(
description=“Return JSON only with keys text and extra_text. Use double quotes. No markdown.”,
expected_output=“Valid JSON object with keys: text, extra_text.”,
agent=agent,
output_pydantic=OutputFormat,
)

3

Thanks for the reply!

Actually, the solution has been to change and run the task instead of the agent. Also, it seems that gpt-3.5-turbo doesn’t allow OutputFormat:

# ---- STATE ----
class State(BaseModel):
    steps: int = 0
    code: str = ""
    llm: str = "gpt-4o-mini"  <--- THIS

llm_instance = ChatOpenAI(
    model_name=self.state.llm,
    temperature=0.7,
)

# Create an Agent
agent_name = Agent(
    role="Friendly assistant",
    goal="Return a message that says 'I worked!'",
    backstory="Testing async agent call",
    tools=[],
    llm=llm_instance,
    verbose=False,
)

print("Starting task execution...")

task = Task(
    description="""Return a JSON object with exactly two fields:
    - 'text': containing the message 'I worked!'
    - 'extra_text': containing 'Success!'""",
    expected_output="JSON object with 'text' and 'extra_text' fields",
    agent=agent_name,
    output_pydantic=OutputFormat,
)

result = task.execute_sync() <--- THIS
4

This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.