Python SDK — LangChain Integration

The recommended way to track LangChain pipelines in production is using the explicit PaygentLangChainCallback. This approach provides fine-grained control over tracking context and enables seamless multi-model pipeline monitoring.

Production Callback Integration

Attach the PaygentLangChainCallback to your model instances to automatically track tokens, costs, and performance across every step of your chain.

langchain_production.py

1 import os
2 import paygent_sdk
3 from paygent_sdk import PaygentLangChainCallback
4 from langchain_openai import ChatOpenAI
5 from langchain_google_genai import ChatGoogleGenerativeAI
6 from langchain_core.prompts import ChatPromptTemplate
7 from langchain_core.output_parsers import StrOutputParser
8 from dotenv import load_dotenv
9 
10 load_dotenv()
11 
12 # 1. Initialize Paygent
13 paygent_sdk.init(os.getenv("PAYGENT_API_KEY"))
14 
15 # 2. Create the production-ready callback
16 # Use this single tracker for the entire pipeline
17 paygent_cb = PaygentLangChainCallback(
18     paygent_client=paygent_sdk._global_client,
19     indicator="article-generator-pipeline",
20     external_agent_id="writer-agent",     # Get this from your Agent settings in Paygent
21     external_customer_id="customer-99",
22     external_customer_name="Global Media" # Optional: Auto-creates customer in Paygent
23 )
24 
25 # 3. Attach to your models
26 llm = ChatOpenAI(
27     model="gpt-4o",
28     temperature=0,
29     callbacks=[paygent_cb]
30 )
31 
32 # 4. Use in a standard LangChain Expression Language (LCEL) chain
33 prompt = ChatPromptTemplate.from_template("Summarize this topic: {topic}")
34 chain = prompt | llm | StrOutputParser()
35 
36 if __name__ == "__main__":
37     result = chain.invoke({"topic": "Quantum Computing"})
38     print(f"Summary: {result}")
39     # ✅ Every invocation is now tracked in your Paygent dashboard

Why use callbacks?

Named Context: Group different parts of your application under distinct indicators.
Auto-Onboarding: Automatically create customers in Paygent when they first interact with your AI.
Multi-Model: Track usage across OpenAI, Anthropic, Gemini, and Mistral in a single unified view.

Multi-Model Pipeline Example

Complex workflows often involve multiple LLM providers. By passing the same callback instance to every model, you maintain a consistent billing context.

multi_model_pipeline.py

1 # Attach the same callback to different providers
2 openai_llm  = ChatOpenAI(model="gpt-4o", callbacks=[paygent_cb])
3 gemini_llm  = ChatGoogleGenerativeAI(model="gemini-2.5-flash", callbacks=[paygent_cb])
4 
5 # Both models will now contribute usage data to the same 'article-generator-pipeline' indicator