respan-instrumentation-agentspec

Respan instrumentation plugin for AgentSpec (pyagentspec).

This package activates the upstream OpenInference AgentSpec span processor through Respan's OpenInference translator, so AgentSpec spans are exported through the same Respan OTLP pipeline used by Respan.

Installation

pip install respan-ai respan-instrumentation-agentspec "pyagentspec[langgraph]"

Usage

from pyagentspec.adapters.langgraph import AgentSpecLoader
from pyagentspec.agent import Agent
from pyagentspec.llms import OpenAiConfig
from respan import Respan
from respan_instrumentation_agentspec import AgentSpecInstrumentor

respan = Respan(
    app_name="agentspec-haiku-agent",
    instrumentations=[
        AgentSpecInstrumentor(workflow_name="agentspec_haiku_agent")
    ],
)

try:
    agent = Agent(
        name="haiku_assistant",
        description="A helpful assistant that writes haikus.",
        llm_config=OpenAiConfig(name="openai", model_id="gpt-4.1-nano"),
        system_prompt="You are a helpful assistant. Respond only with a haiku.",
    )

    langgraph_agent = AgentSpecLoader().load_component(agent)
    result = langgraph_agent.invoke(
        input={"messages": [{"role": "user", "content": "Write a haiku about tracing."}]}
    )

    print(result["messages"][-1].content)
finally:
    respan.shutdown()
    respan.flush()