Langchain (Python)

Portkey adds core production capabilities to any Langchain app.

This guide covers the integration for the Python flavour of Langchain. Docs for the JS Langchain integration are here.

LangChain is a framework for developing applications powered by language models. It enables applications that:

• Are context-aware: connect a language model to sources of context (prompt instructions, few shot examples, content to ground its response in, etc.)

• Reason: rely on a language model to reason (about how to answer based on provided context, what actions to take, etc.)

You can find more information about it here.

When using Langchain, Portkey can help take it to production by adding a fast AI gateway, observability, prompt management and more to your Langchain app.

Quick Start Integration

Install the Portkey and Langchain Python SDKs to get started.

pip install -U langchain-core portkey_ai langchain-openai

We installed langchain-core to skip the optional dependencies. You can also install langchain if you prefer.

Since Portkey is fully compatible with the OpenAI signature, you can connect to the Portkey Ai Gateway through the ChatOpenAI interface.

• Set the base_url as PORTKEY_GATEWAY_URL

• Add default_headers to consume the headers needed by Portkey using the createHeaders helper method.

We can now initialise the model and update the model to use Portkey's AI gateway

from langchain_openai import ChatOpenAI
from portkey_ai import createHeaders, PORTKEY_GATEWAY_URL

PORTKEY_API_KEY = "..."
PROVIDER_API_KEY = "..." # Add the API key of the AI provider being used

portkey_headers = createHeaders(api_key=PORTKEY_API_KEY,provider="openai")

llm = ChatOpenAI(api_key=PROVIDER_API_KEY, base_url=PORTKEY_GATEWAY_URL, default_headers=portkey_headers)

llm.invoke("What is the meaning of life, universe and everything?")

Response

AIMessage(content='The meaning of life, universe, and everything is a question that has puzzled humanity for centuries. In the book "The Hitchhiker\'s Guide to the Galaxy" by Douglas Adams, the answer to this ultimate question is humorously given as "42." This has since become a popular meme and cultural reference.\n\nIn a more philosophical sense, the meaning of life, universe, and everything is subjective and can vary greatly from person to person. Some may find meaning in religious beliefs, others in personal relationships, achievements, or experiences. Ultimately, it is up to each individual to find their own meaning and purpose in life.', response_metadata={'token_usage': {'completion_tokens': 124, 'prompt_tokens': 18, 'total_tokens': 142}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': 'fp_c2295e73ad', 'finish_reason': 'stop', 'logprobs': None}, id='run-d2b790be-589b-4c72-add8-a36e098ab277-0')

The call and the corresponding prompt will also be visible on the Portkey logs tab.

Using Virtual Keys for Multiple Models

Portkey supports Virtual Keys which are an easy way to store and manage API keys in a secure vault. Lets try using a Virtual Key to make LLM calls.

1. Create a Virtual Key in your Portkey account and copy the id

Let's try creating a new Virtual Key for Mistral like this

2. Use Virtual Keys in the Portkey Headers

The virtual_key parameter sets the authentication and provider for the AI provider being used. In our case we're using the Mistral Virtual key.

Notice that the api_key can be left blank as that authentication won't be used.

from langchain_openai import ChatOpenAI
from portkey_ai import createHeaders, PORTKEY_GATEWAY_URL

PORTKEY_API_KEY = "..."
VIRTUAL_KEY = "..." # Mistral's virtual key we copied above

portkey_headers = createHeaders(api_key=PORTKEY_API_KEY,virtual_key=VIRTUAL_KEY)

llm = ChatOpenAI(api_key="X", base_url=PORTKEY_GATEWAY_URL, default_headers=portkey_headers, model="mistral-large-latest")

llm.invoke("What is the meaning of life, universe and everything?")

The Portkey AI gateway will authenticate the API request to Mistral and get the response back in the OpenAI format for you to consume.

The AI gateway extends Langchain's ChatOpenAI class making it a single interface to call any provider and any model.

Embeddings

Embeddings in Langchain through Portkey work the same way as the Chat Models using the OpenAIEmbeddings class. Let's try to create an embedding using OpenAI's embedding model

from langchain_openai import OpenAIEmbeddings  

PORTKEY_API_KEY = "..."
VIRTUAL_KEY = "..." # Add OpenAI's Virtual Key created in Portkey

portkey_headers = createHeaders(api_key=PORTKEY_API_KEY,virtual_key=VIRTUAL_KEY)
  
embeddings_model = OpenAIEmbeddings(api_key="X", base_url=PORTKEY_GATEWAY_URL, default_headers=portkey_headers)
embeddings = embeddings_model.embed_documents(["Hi there!"])
len(embeddings[0])

Only OpenAI is supported as an embedding provider for Langchain because internally, Langchain converts the texts into tokens which are then sent as input to the API. This method of embedding tokens instead of strings via the API is ONLY supported by OpenAI.

If you plan to use any other embedding model, we recommend using the Portkey SDK directly to make embedding calls.

Chains & Prompts

Chains enable the integration of various Langchain concepts for simultaneous execution while Langchain supports Prompt Templates to construct inputs for language models. Lets see how this would work with Portkey

from langchain_openai import ChatOpenAI
from portkey_ai import createHeaders, PORTKEY_GATEWAY_URL
from langchain_core.prompts import ChatPromptTemplate

PORTKEY_API_KEY = "..."
VIRTUAL_KEY = "..." # Add the OpenAI Virtual key

portkey_headers = createHeaders(api_key=PORTKEY_API_KEY,provider="openai")

chat = ChatOpenAI(api_key=PROVIDER_API_KEY, base_url=PORTKEY_GATEWAY_URL, default_headers=portkey_headers)

prompt = ChatPromptTemplate.from_messages([ 
	("system", "You are world class technical documentation writer."), 
	("user", "{input}") 
])
  
chain = prompt | chat 
chain.invoke({"input": "Explain the concept of an API"})

We'd be able to view the exact prompt that was used to make the call to OpenAI in the Portkey logs dashboards.

Using Advanced Routing

The Portkey AI Gateway brings capabilities like load-balancing, fallbacks, experimentation and canary testing to Langchain through a configuration-first approach.

Let's take an example where we might want to split traffic between gpt-4 and claude-opus 50:50 to test the two large models. The gateway configuration for this would look like the following:

config = {
    "strategy": {
         "mode": "loadbalance"
    },
    "targets": [{
        "virtual_key": "openai-25654", # OpenAI's virtual key
        "override_params": {"model": "gpt4"},
        "weight": 0.5
    }, {
        "virtual_key": "anthropic-25654", # Anthropic's virtual key
        "override_params": {"model": "claude-3-opus-20240229"},
        "weight": 0.5
    }]
}

We can then use this config in our requests being made from langchain.

portkey_headers = createHeaders(
    api_key=PORTKEY_API_KEY,
    config=config
)

llm = ChatOpenAI(api_key="X", base_url=PORTKEY_GATEWAY_URL, default_headers=portkey_headers)

llm.invoke("What is the meaning of life, universe and everything?")

When the LLM is invoked, Portkey will distribute the requests to gpt-4 and claude-3-opus-20240229 in the ratio of the defined weights.

You can find more config examples here.

Agents & Tracing

A powerful capability of Langchain is creating Agents. The challenge with agentic workflows is that prompts are often abstracted out and it's hard to get a visibility into what the agent is doing. This also makes debugging harder.

Portkey's Langchain integration gives you full visibility into the running of an agent. Let's take an example of a popular agentic workflow.

"""
Libraries:
    pip install langchain langchain-openai langchainhub portkey-ai
"""

import os

from langchain import hub
from langchain.agents import AgentExecutor, create_openai_tools_agent, tool
from langchain_openai import ChatOpenAI
from portkey_ai import PORTKEY_GATEWAY_URL, createHeaders

prompt = hub.pull("hwchase17/openai-tools-agent")

portkey_headers = createHeaders(
    api_key=os.getenv("PORTKEY_API_KEY"),
    config="your config id",
    trace_id="uuid-uuid-uuid-uuid",
)


@tool
def add(first_int: int, second_int: int) -> int:
    return first_int + second_int


@tool
def multiply(first_int: int, second_int: int) -> int:
    return first_int * second_int


@tool
def exponentiate(base: int, exponent: int) -> int:
    return base**exponent


tools = [multiply, add, exponentiate]

model = ChatOpenAI(
    api_key="ignore",  # type: ignore
    base_url=PORTKEY_GATEWAY_URL,
    default_headers=portkey_headers,
    temperature=0,
)

# Construct the OpenAI Tools agent
agent = create_openai_tools_agent(model, tools, prompt)  # type: ignore

# Create an agent executor by passing in the agent and tools
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)  # type: ignore

agent_executor.invoke(
    {
        "input": "Take 3 to the fifth power and multiply that by the sum of twelve and three, then square the whole result"
    }
)

Running this would yield the following logs in Portkey.

This is extremely powerful since you gain control and visibility over the agent flows so you can identify problems and make updates as needed.