agentinsight 0.1.0a1

Python observability SDK for agentic systems

AgentInsight Python SDK

Drop-in observability for agentic systems. The SDK records agent, LLM, tool, retrieval, orchestration, and custom spans, then exports them locally or to an AgentInsight backend.

Install

pip install agentinsight

For local development from this repository:

pip install -e sdk

Quick Start

import agentinsight
import openai

agentinsight.init(
    service_name="support-agent",
    exporter="cloud",
    endpoint="https://your-agentinsight-app.vercel.app",
    api_key="ai_...",
)

client = openai.OpenAI()
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}],
)

OpenAI and Anthropic calls are automatically traced when those packages are installed. Use @observe and trace_context for custom agent logic.

Local Export

import agentinsight
from agentinsight import observe

agentinsight.init(exporter="local", local_path="agentinsight-spans.jsonl")

@observe(kind="agent")
def run_agent(task: str):
    return {"answer": task.upper()}

run_agent("triage this incident")

Manual Spans

from agentinsight import trace_context

with trace_context("plan_trip", kind="agent", attributes={"agent.name": "planner"}) as span:
    span.set_attribute("agent.step", "plan")

LangChain / LangGraph

from agentinsight.integrations.langchain import AgentInsightCallbackHandler

handler = AgentInsightCallbackHandler()
chain.invoke({"input": "hello"}, config={"callbacks": [handler]})

LangGraph uses LangChain callbacks, so the same handler works for many LangGraph apps.