phospho can be added to a Langchain agent as a callback handler. By default, the task input is the beginning of the chain, and the task output is the end result. Intermediate steps are also logged.

from phospho.integrations import PhosphoLangchainCallbackHandler

chain = ... # Your Langchain agent or chain

chain.invoke(
    "Your chain input",
    # Add the callback handler to the config
    config={"callbacks": [PhosphoLangchainCallbackHandler()]},
)

​
Detailed setup in a retrieval agent

1

Set the following environment variables:

export PHOSPHO_API_KEY=...
export PHOSPHO_PROJECT_ID=...
export OPENAI_API_KEY=...

Learn how to get your project id and api key by clicking here!

2

Install requirements:

pip install phospho openai langchain faiss-cpu

The phospho module implements the Langchain callback as well as other helpful tools to interact with phospho. Learn more in the python doc.

The phospho module is an open source work in progress. Your help is deeply appreciated!

3

Create a file called main.py with the agent code.

phospho is integrated with langchain via the PhosphoLangchainCallbackHandler callback handler. This callback handler will log the input and output of the agent to phospho.


from langchain.prompts import ChatPromptTemplate
from langchain_community.chat_models import ChatOpenAI
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough

vectorstore = FAISS.from_texts(
    [
        "phospho is the LLM analytics platform",
        "Paris is the capital of Fashion (sorry not sorry London)",
        "The Concorde had a maximum cruising speed of 2,179 km (1,354 miles) per hour, or Mach 2.04 (more than twice the speed of sound), allowing the aircraft to reduce the flight time between London and New York to about three hours.",
    ],
    embedding=OpenAIEmbeddings(),
)
retriever = vectorstore.as_retriever()
template = """Answer the question based only on the following context:
{context}

Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)
model = ChatOpenAI()

retrieval_chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | model
    | StrOutputParser()
)


# To integrate with Phospho, add the following callback handler

from phospho.integrations import PhosphoLangchainCallbackHandler


while True:
    text = input("Enter a question: ")
    response = retrieval_chain.invoke(
        text, 
        config={
            "callbacks": [PhosphoLangchainCallbackHandler()]
        }
    )
    print(response)

The integration with phospho is done by adding the PhosphoLangchainCallbackHandler to the config of the chain. You can learn more about callbacks in the langchain doc.

4

Start the RAG agent and ask questions about the documents.

python main.py

The agent answers question based on retrieved documents (RAG, Retrieval Augmented Generation).

Enter a question: What's the top speed of the Concorde?
The Concorde top speed is 2,179km per hour.

The conversation and the intermediate retrievals steps (such as the documents retrieved) are logged to phospho.

​
Custom logging in langchain

For more advanced manual logging with a langchain, you can inherit from the PhosphoLangchainCallbackHandler and add custom behaviour.

The callback has a reference to the phospho object, which can be used to log custom data.

from phospho.integrations import PhosphoLangchainCallbackHandler

class MyCustomLangchainCallbackHandler(PhosphoLangchainCallbackHandler):

    def on_agent_finish(self, finish: AgentFinish, **kwargs: Any) -> Any:
        """Run on agent end."""

        # Do something custom here
        self.phospho.log(input="...", output="...")

You can refer to the langchain doc to have the full list of callbacks available.

Was this page helpful?

Log to phospho with JavascriptImport from Langsmith 🦜🔗