Skip to main content

APIs

Introduction​

APIChain allows using large language models (LLMs) to interact with APIs by constructing a chain of the LLM and API docs. By providing a natural language question relevant to the API documentation, APIChain can query the API and retrieve information.

This reference guide will cover:

  • Basic usage and setup
  • Examples with OpenMeteo, TMDB, ListenNotes and more
  • Advanced querying and chaining
  • Customizing prompts and parsing

Basic Usage​

To create an APIChain, you need:

  • An LLM like OpenAI
  • API documentation
  • Optional headers if authentication required
from langchain.chains import APIChain
from langchain.llms import OpenAI

llm = OpenAI(temperature=0)

chain = APIChain.from_llm_and_api_docs(llm, api_docs, headers=headers)

Construct the chain by passing the LLM, API docs, and optional headers.

Then make queries by running the chain:

response = chain.run("What is the weather in London tomorrow?")

Examples​

OpenMeteo​

from langchain.chains.api import open_meteo_docs

chain = APIChain.from_llm_and_api_docs(llm, open_meteo_docs.OPEN_METEO_DOCS, verbose=True)

response = chain.run('What is the weather like right now in Munich, Germany in degrees Fahrenheit?')

This initializes a chain with the OpenMeteo docs. The query asks for the current weather in Munich in Fahrenheit.

TMDB​

import os
os.environ['TMDB_API_KEY'] = "xxx"

from langchain.chains.api import tmdb_docs
headers = {"Authorization": f"Bearer {os.environ['TMDB_API_KEY']}"}

chain = APIChain.from_llm_and_api_docs(llm, tmdb_docs.TMDB_DOCS, headers=headers)

response = chain.run("Search for movies about 'Artificial Intelligence'")

This shows using the TMDB API by passing the API key in headers. The chain is constructed with the TMDB docs and used to search for AI movies.

Listen Notes​

import os
listen_api_key = os.environ["LISTEN_API_KEY"]
headers = {"X-ListenAPI-Key": listen_api_key}

from langchain.chains.api import podcast_docs

chain = APIChain.from_llm_and_api_docs(llm, podcast_docs.PODCAST_DOCS, headers=headers)

response = chain.run("Find podcasts about meditation")

This initializes a chain with the Listen Notes podcast API, passing the API key. It searches for meditation podcasts.

Advanced Usage​

The examples above show basic single queries. APIChain also supports:

  • Chaining multiple APIs by composing chains
  • Advanced response parsing with Python
  • Caching responses
  • Custom prompts for each API
  • Error handling

Chaining Multiple APIs​

Chains for different APIs can be composed:

open_meteo_chain = APIChain(llm, open_meteo_docs)
tmdb_chain = APIChain(llm, tmdb_docs, headers)

composed_chain = open_meteo_chain + tmdb_chain

response = composed_chain.run("What's the weather tomorrow in Los Angeles? What are popular movies set in LA?")

This composes an OpenMeteo and TMDB chain, allowing querying both APIs.

Response Parsing​

API responses can be parsed in Python before returning:

def parse_response(response):
    data = json.loads(response)
    return data["title"]

chain = APIChain(llm, api_docs, response_parser=parse_response)

title = chain.run("Get title of most popular movie on TMDB")

A response parser function can extract the relevant data.

Custom Prompts​

The prompt can be customized per API:

prompt = PromptTemplate(
    input="Question: {question}",
    api_docs=api_docs,
    output="Response: {response}"
)

chain = APIChain(llm, prompt=prompt)

This allows modifying the prompt structure.

Conclusion​

In summary, APIChain enables seamless integration of LLMs with APIs using natural language. The examples here covered common use cases and customization techniques. With APIChain, you can leverage the knowledge of large models to query and compose APIs.