Gmail

This notebook walks through connecting a LangChain email to the Gmail API.

To use this toolkit, you will need to set up your credentials explained in the Gmail API docs. Once you've downloaded the credentials.json file, you can start using the Gmail API. Once this is done, we'll install the required libraries.

%pip install --upgrade --quiet  google-api-python-client > /dev/null
%pip install --upgrade --quiet  google-auth-oauthlib > /dev/null
%pip install --upgrade --quiet  google-auth-httplib2 > /dev/null
%pip install --upgrade --quiet  beautifulsoup4 > /dev/null # This is optional but is useful for parsing HTML messages

You also need to install the langchain-community package where the integration lives:

pip install -U langchain-community

It's also helpful (but not needed) to set up LangSmith for best-in-class observability

# os.environ["LANGCHAIN_TRACING_V2"] = "true"
# os.environ["LANGCHAIN_API_KEY"] = getpass.getpass()

Create the Toolkit

By default the toolkit reads the local credentials.json file. You can also manually provide a Credentials object.

from langchain_community.agent_toolkits import GmailToolkit

toolkit = GmailToolkit()

Customizing Authentication

Behind the scenes, a googleapi resource is created using the following methods. you can manually build a googleapi resource for more auth control.

from langchain_community.tools.gmail.utils import (
    build_resource_service,
    get_gmail_credentials,
)

# Can review scopes here https://developers.google.com/gmail/api/auth/scopes
# For instance, readonly scope is 'https://www.googleapis.com/auth/gmail.readonly'
credentials = get_gmail_credentials(
    token_file="token.json",
    scopes=["https://mail.google.com/"],
    client_secrets_file="credentials.json",
)
api_resource = build_resource_service(credentials=credentials)
toolkit = GmailToolkit(api_resource=api_resource)

tools = toolkit.get_tools()
tools

[GmailCreateDraft(name='create_gmail_draft', description='Use this tool to create a draft email with the provided message fields.', args_schema=<class 'langchain_community.tools.gmail.create_draft.CreateDraftSchema'>, return_direct=False, verbose=False, callbacks=None, callback_manager=None, api_resource=<googleapiclient.discovery.Resource object at 0x10e5c6d10>),
 GmailSendMessage(name='send_gmail_message', description='Use this tool to send email messages. The input is the message, recipents', args_schema=None, return_direct=False, verbose=False, callbacks=None, callback_manager=None, api_resource=<googleapiclient.discovery.Resource object at 0x10e5c6d10>),
 GmailSearch(name='search_gmail', description=('Use this tool to search for email messages or threads. The input must be a valid Gmail query. The output is a JSON list of the requested resource.',), args_schema=<class 'langchain_community.tools.gmail.search.SearchArgsSchema'>, return_direct=False, verbose=False, callbacks=None, callback_manager=None, api_resource=<googleapiclient.discovery.Resource object at 0x10e5c6d10>),
 GmailGetMessage(name='get_gmail_message', description='Use this tool to fetch an email by message ID. Returns the thread ID, snipet, body, subject, and sender.', args_schema=<class 'langchain_community.tools.gmail.get_message.SearchArgsSchema'>, return_direct=False, verbose=False, callbacks=None, callback_manager=None, api_resource=<googleapiclient.discovery.Resource object at 0x10e5c6d10>),
 GmailGetThread(name='get_gmail_thread', description=('Use this tool to search for email messages. The input must be a valid Gmail query. The output is a JSON list of messages.',), args_schema=<class 'langchain_community.tools.gmail.get_thread.GetThreadSchema'>, return_direct=False, verbose=False, callbacks=None, callback_manager=None, api_resource=<googleapiclient.discovery.Resource object at 0x10e5c6d10>)]

Usage

We show here how to use it as part of an agent. We use the OpenAI Functions Agent, so we will need to setup and install the required dependencies for that. We will also use LangSmith Hub to pull the prompt from, so we will need to install that.

pip install -U langchain-openai langchainhub

import getpass
import os

os.environ["OPENAI_API_KEY"] = getpass.getpass()

from langchain import hub
from langchain.agents import AgentExecutor, create_openai_functions_agent
from langchain_openai import ChatOpenAI

instructions = """You are an assistant."""
base_prompt = hub.pull("langchain-ai/openai-functions-template")
prompt = base_prompt.partial(instructions=instructions)

llm = ChatOpenAI(temperature=0)

agent = create_openai_functions_agent(llm, toolkit.get_tools(), prompt)

agent_executor = AgentExecutor(
    agent=agent,
    tools=toolkit.get_tools(),
    # This is set to False to prevent information about my email showing up on the screen
    # Normally, it is helpful to have it set to True however.
    verbose=False,
)

agent_executor.invoke(
    {
        "input": "Create a gmail draft for me to edit of a letter from the perspective of a sentient parrot"
        " who is looking to collaborate on some research with her"
        " estranged friend, a cat. Under no circumstances may you send the message, however."
    }
)

{'input': 'Create a gmail draft for me to edit of a letter from the perspective of a sentient parrot who is looking to collaborate on some research with her estranged friend, a cat. Under no circumstances may you send the message, however.',
 'output': 'I have created a draft email for you to edit. Please find the draft in your Gmail drafts folder. Remember, under no circumstances should you send the message.'}

agent_executor.invoke(
    {"input": "Could you search in my drafts for the latest email? what is the title?"}
)

{'input': 'Could you search in my drafts for the latest email? what is the title?',
 'output': 'The latest email in your drafts is titled "Collaborative Research Proposal".'}