Question about observability tool integration

[boyangsvl]

Discussed in #321

^{Originally posted by sr198 April 21, 2025}
Going through the Opik integration example and couples of things stand out.

opik_tracer = OpikTracer()

root_agent = Agent(
name="weather_time_agent",
model="gemini-2.0-flash-exp",
description=("Agent to answer questions about the time and weather in a city."),
instruction=("I can answer your questions about the time and weather in a city."),
tools=[get_weather, get_current_time],
before_agent_callback=opik_tracer.before_agent_callback,
after_agent_callback=opik_tracer.after_agent_callback,
before_model_callback=opik_tracer.before_model_callback,
after_model_callback=opik_tracer.after_model_callback,
before_tool_callback=opik_tracer.before_tool_callback,
after_tool_callback=opik_tracer.after_tool_callback,
)

1. The callback params take a single value today - which means, if I want to add my own callback for some guardrail feature then I will have to wrap OpikTracker in my code. Should we allow callbacks to accept lists for chained execution at the framework level?
   Agent(
   ...
   before_model_callback=[my_callback, opik_tracer.before_model_callback],
   )
2. Also can we allow observability integration at a higher/global level, so it’s not repeated per agent.

What do you think?

[stefanadelbert]

I'm struggling with a similar problem. I was using a before_model_callback and then applied Opik tracing only to realise that I had blatted over my callback handler. I believe it's possible to provide a list of callback handlers to some of the callbacks, but not all. There is code in LlmAgent, e.g. canonical_before_model_callbacks, which handles the case where before_model_callback is already a list. But if you want to apply opik tracing selectively, you need to be careful to not overwrite existing callbacks.
Also, it's really painful to apply opik tracing to a mulit-agent as you have to set the callbacks for each and every agent. And the callbacks to set depend on the type of the agent, e.g. BaseAgent only has before_agent_callback and after_agent_callback (which can be lists).
While the use of callbacks is convenient, the implementation seems clunky.

[xuanyang15]

We have updated callbacks to also accept a list and added plugins on the runner level.