agentrace-ai 0.1.0

Lightweight OpenTelemetry-native tracing for LLM and AI agent applications

agentrace-ai

Lightweight OpenTelemetry-native tracing for LLM and AI agent applications.

pip install agentrace-ai

Quick Start

import agentrace
from agentrace import trace_llm, trace_tool, trace_agent

agentrace.init(service_name="my-agent")

@trace_tool(tool_name="search")
def search(query):
    return ["result1", "result2"]

@trace_llm(model="gpt-4o")
def call_llm(messages):
    return client.chat.completions.create(model="gpt-4o", messages=messages)

@trace_agent(name="research-agent")
def run(question):
    results = search(question)
    return call_llm([{"role": "user", "content": str(results)}])

run("What is OpenTelemetry?")
agentrace.shutdown()

Features

• Decorators: @observe, @trace_llm, @trace_tool, @trace_agent, @trace_chain, @trace_retrieval
• Context Managers: trace_span(), trace_llm_call() with .record_response()
• Session Tracking: set_session(), set_user(), session_context()
• Cost Tracking: Auto-calculates cost from token usage for OpenAI and Anthropic models
• Auto-Instrumentation: Monkey-patch OpenAI/Anthropic SDKs with one line
• LangChain Callback: AgentraceCallbackHandler for LangChain/LangGraph
• Export Anywhere: Console, OTLP (gRPC/HTTP), or build your own exporter
• Lightweight: Only 2 dependencies (opentelemetry-api + opentelemetry-sdk)

Integrations

pip install agentrace-ai[openai]      # OpenAI auto-instrumentation
pip install agentrace-ai[anthropic]   # Anthropic auto-instrumentation
pip install agentrace-ai[langchain]   # LangChain callback handler
pip install agentrace-ai[otlp]       # OTLP exporter (Jaeger, Grafana, etc.)
pip install agentrace-ai[all]        # Everything

Auto-instrument OpenAI

from agentrace.integrations import openai_patch
openai_patch.instrument()
# All OpenAI calls are now traced automatically

Auto-instrument Anthropic

from agentrace.integrations import anthropic_patch
anthropic_patch.instrument()

LangChain / LangGraph

from agentrace.integrations.langchain_cb import AgentraceCallbackHandler
chain.invoke({"input": "..."}, config={"callbacks": [AgentraceCallbackHandler()]})

Export to OTLP (Jaeger, Grafana Tempo, etc.)

agentrace.init(
    service_name="my-agent",
    exporters=["console", "otlp"],
    otlp_endpoint="http://localhost:4317",
)

License

MIT