Instrument

Add observability to your AI agents using OpenTelemetry. AgentHealth captures traces, tool calls, token usage, and more.

Option A: Auto-Instrumentation

For supported frameworks, just initialize the SDK:

from opensearch_agenthealth import AgentHealth

# Initialize - automatic instrumentation begins
agenthealth = AgentHealth(
    endpoint="http://localhost:4317",  # OTEL Collector
    service_name="my-agent"
)

# Your agent code runs as normal - traces are captured automatically
result = my_agent.run("Analyze this error log")

Option B: Manual Instrumentation

For custom agents, wrap your code with trace contexts:

from opentelemetry import trace

tracer = trace.get_tracer("my-agent")

with tracer.start_as_current_span("agent_task") as span:
    # Add attributes
    span.set_attribute("gen_ai.request.model", "claude-sonnet-4")

    # Your agent logic
    response = llm.generate(prompt)

    # Record token usage
    span.set_attribute("gen_ai.usage.input_tokens", response.input_tokens)
    span.set_attribute("gen_ai.usage.output_tokens", response.output_tokens)

Next Steps

Full Instrumentation Guide →