Hard coded openai URL in processors.py

[mariansollmann]

I think this URL should be flexible and support other APIs

src/agents/tracing/processors.py
line 36
endpoint: str = "https://api.openai.com/v1/traces/ingest",

[seratch]

Hi, thanks for writing in.

Indeed, the default string is static but you can instantiate BatchTraceProcessor along with BackendSpanExporter, and then replace the default one by passing your own ones to set_trace_processors(): https://openai.github.io/openai-agents-python/tracing/#custom-tracing-processors

Is there any scenario that the above customization does not work?

[mariansollmann]

Thank you very much for getting back to me. I just noticed. I'm clearly doing something wrong, could you give a simple example like this one and showcase how I could use another API?

from agents import Agent, Runner

agent = Agent(name="Assistant", instructions="You are a helpful assistant")

result = Runner.run_sync(agent, "Write a haiku about recursion in programming.")
print(result.final_output)

[mariansollmann]

I got it running by instantiating my client like this

custom_client = AsyncOpenAI(
    base_url="https://myurl.com",
    api_key=os.getenv("MY_API_KEY")
)
set_default_openai_client(custom_client)

I still get an "openai.agents:[non-fatal] Tracing client error 404" due to the wrong tracing URL

[ihower]

Here is a minimal code to replace the Tracing Processor:

from agents.tracing import TracingProcessor

class CustomProcessor(TracingProcessor):
    def __init__(self):
        self.active_traces = {}
        self.active_spans = {}

    def on_trace_start(self, trace):
        print(f"Trace started: {trace}")
        self.active_traces[trace.trace_id] = trace

    def on_trace_end(self, trace):
        print(f"Trace ended: {trace}")
        del self.active_traces[trace.trace_id]

    def on_span_start(self, span):
        print(f"Span started: {span}")
        self.active_spans[span.span_id] = span

    def on_span_end(self, span):
        print(f"Span ended: {span}")
        # Process completed span
        del self.active_spans[span.span_id]

    def shutdown(self):
        print("Shutting down processor...")
        # Clean up resources
        self.active_traces.clear()
        self.active_spans.clear()

    def force_flush(self):
        print("Force flushing processor...")
    
from agents import Agent, Runner, set_trace_processors
 
set_trace_processors([CustomProcessor()])

agent = Agent(
  name="Assistant",
  instructions="Just say hello world",
)

result = Runner.run_sync(agent, "Hi")
print(result.final_output)

The Processor interface documentation is here:
https://openai.github.io/openai-agents-python/ref/tracing/processor_interface/

For example, here is Braintrust’s implementation:
https://github.com/braintrustdata/braintrust-sdk/blob/main/py/src/braintrust/wrappers/openai.py

And here is Langsmith’s implementation:
https://github.com/langchain-ai/langsmith-sdk/blob/main/python/langsmith/wrappers/_openai_agents.py

Most third-party monitoring platforms have integrations for the OpenAI Agents SDK, so you don’t need to implement it yourself.

[ihower]

@mariansollmann The base_url in your code refers to the LLM endpoint, not tracing.
Tracing Processor is not just a URL endpoint, but more like a hook. You need to implement in the hook where the trace should be sent and recorded.

[mariansollmann]

Thank you very much for the great example and explanation. Last question: Shouldn't all be disabled and no more warnings when I add OPENAI_AGENTS_DISABLE_TRACING=1 to my .env?

[ihower]

@mariansollmann Setting OPENAI_AGENTS_DISABLE_TRACING=1 should disable all tracing, including both OpenAI and custom providers, without warnings.

[github-actions]

This issue is stale because it has been open for 7 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 3 days since being marked as stale.