1+ """
2+ PostHog LangChain Integration Example
3+
4+ This example demonstrates how to use PostHog tools with LangChain using
5+ the local posthog_agent_toolkit package. It shows how to analyze product
6+ usage data similar to the TypeScript example.
7+ """
8+
9+ import asyncio
10+ import os
11+ import sys
12+
13+ from dotenv import load_dotenv
14+ from langchain_openai import ChatOpenAI
15+ from langchain .agents import AgentExecutor , create_tool_calling_agent
16+ from langchain_core .prompts import ChatPromptTemplate , MessagesPlaceholder
17+ from posthog_agent_toolkit .integrations .langchain .toolkit import PostHogAgentToolkit
18+
19+
20+ async def analyze_product_usage ():
21+ """Analyze product usage using PostHog data."""
22+
23+ print ("🚀 PostHog LangChain Agent - Product Usage Analysis\n " )
24+
25+ # Initialize the PostHog toolkit with credentials
26+ toolkit = PostHogAgentToolkit (
27+ personal_api_key = os .getenv ("POSTHOG_PERSONAL_API_KEY" ),
28+ url = os .getenv ("POSTHOG_MCP_URL" , "https://mcp.posthog.com/mcp" )
29+ )
30+
31+ # Get the tools
32+ tools = await toolkit .get_tools ()
33+
34+ # Initialize the LLM
35+ llm = ChatOpenAI (
36+ model = "gpt-4o-mini" ,
37+ temperature = 0 ,
38+ api_key = os .getenv ("OPENAI_API_KEY" )
39+ )
40+
41+ # Create a system prompt for the agent
42+ prompt = ChatPromptTemplate .from_messages ([
43+ (
44+ "system" ,
45+ "You are a data analyst. Your task is to do a deep dive into what's happening in our product. "
46+ "Be concise and data-driven in your responses."
47+ ),
48+ ("human" , "{input}" ),
49+ MessagesPlaceholder ("agent_scratchpad" ),
50+ ])
51+
52+ agent = create_tool_calling_agent (
53+ llm = llm ,
54+ tools = tools ,
55+ prompt = prompt ,
56+ )
57+
58+ agent_executor = AgentExecutor (
59+ agent = agent ,
60+ tools = tools ,
61+ verbose = False ,
62+ max_iterations = 5 ,
63+ )
64+
65+ # Invoke the agent with an analysis request
66+ result = await agent_executor .ainvoke ({
67+ "input" : """Please analyze our product usage:
68+
69+ 1. Get all available insights (limit 100)
70+ 2. Pick the 5 MOST INTERESTING and VALUABLE insights - prioritize:
71+ - User behavior and engagement metrics
72+ - Conversion funnels
73+ - Retention and growth metrics
74+ - Product adoption insights
75+ - Revenue or business KPIs
76+ AVOID picking feature flag insights unless they show significant business impact
77+ 3. For each selected insight, query its data and explain why it's important
78+ 4. Summarize the key findings in a brief report with actionable recommendations
79+
80+ Focus on insights that tell a story about user behavior and business performance."""
81+ })
82+
83+ print ("\n 📊 Analysis Complete!\n " )
84+ print ("=" * 50 )
85+ print (result ["output" ])
86+ print ("=" * 50 )
87+
88+
89+ async def main ():
90+ """Main function to run the product usage analysis."""
91+ try :
92+ # Load environment variables
93+ load_dotenv ()
94+
95+ # Run the analysis
96+ await analyze_product_usage ()
97+ except Exception as error :
98+ print (f"Error: { error } )
99+ sys .exit (1 )
100+
101+
102+ if __name__ == "__main__" :
103+ asyncio .run (main ())
0 commit comments