Executive AI Assistant

Executive AI Assistant (EAIA) is an AI agent that attempts to do the job of an Executive Assistant (EA).

For a hosted version of EAIA, see documentation here.

Table of contents

• General Setup
  • Env
  • Credentials
  • Configuration
• Run locally
  • Setup EAIA
  • Ingest emails
  • Connect to Agent Inbox
  • Use Agent Inbox
• Run in production (LangGraph Cloud)
  • Setup EAIA on LangGraph Cloud
  • Ingest manually
  • Set up cron job

General Setup

Env

1. Fork and then clone this repo. Note: make sure to fork it first, as in order to deploy this you will need your own repo.
2. Create a Python virtualenv and activate it (e.g. pyenv virtualenv 3.11.1 eaia, pyenv activate eaia)
3. Run pip install -e . to install dependencies and the package

Set up credentials

1. Export OpenAI API key (export OPENAI_API_KEY=...)
2. Export Anthropic API key (export ANTHROPIC_API_KEY=...)
3. Enable Google
   1. Enable the API
      • Enable Gmail API if not already by clicking the blue button Enable the API
   2. Authorize credentials for a desktop application
      1. Download the client secret. After that, run these commands:
      2. mkdir eaia/.secrets - This will create a folder for secrets
      3. mv ${PATH-TO-CLIENT-SECRET.JSON} eaia/.secrets/secrets.json - This will move the client secret you just created to that secrets folder
      4. python scripts/setup_gmail.py - This will generate another file at eaia/.secrets/token.json for accessing Google services.
4. Export LangSmith API key (export LANGSMITH_API_KEY)

Configuration

The configuration for EAIA can be found in eaia/main/config.yaml. Every key in there is required. These are the configuration options:

• email: Email to monitor and send emails as. This should match the credentials you loaded above.
• full_name: Full name of user
• name: First name of user
• background: Basic info on who the user is
• timezone: Default timezone the user is in
• schedule_preferences: Any preferences for how calendar meetings are scheduled. E.g. length, name of meetings, etc
• background_preferences: Any background information that may be needed when responding to emails. E.g. coworkers to loop in, etc.
• response_preferences: Any preferences for what information to include in emails. E.g. whether to send calendly links, etc.
• rewrite_preferences: Any preferences for the tone of your emails
• triage_no: Guidelines for when emails should be ignored
• triage_notify: Guidelines for when user should be notified of emails (but EAIA should not attempt to draft a response)
• triage_email: Guidelines for when EAIA should try to draft a response to an email

Run locally

You can run EAIA locally. This is useful for testing it out, but when wanting to use it for real you will need to have it always running (to run the cron job to check for emails). See this section for instructions on how to run in production (on LangGraph Cloud)

Set up EAIA locally

1. Install development server pip install -U "langgraph-cli[inmem]"
2. Run development server langgraph dev

Ingest Emails Locally

Let's now kick off an ingest job to ingest some emails and run them through our local EAIA.

Leave the langgraph dev command running, and open a new terminal. From there, get back into this directory and virtual environment. To kick off an ingest job, run:

python scripts/run_ingest.py --minutes-since 120 --rerun 1 --early 0

This will ingest all emails in the last 120 minutes (--minutes-since). It will NOT break early if it sees an email it already saw (--early 0) and it will rerun ones it has seen before (--rerun 1). It will run against the local instance we have running.

Set up Agent Inbox with Local EAIA

After we have run it locally, we can interract with any results.

1. Go to Agent Inbox
2. Connect this to your locally running EAIA agent:
   1. Click into Settings
   2. Input your LangSmith API key.
   3. Click Add Inbox
      1. Set Assistant/Graph ID to main
      2. Set Deployment URL to http://127.0.0.1:2024
      3. Give it a name like Local EAIA
      4. Press Submit

You can now interract with EAIA in the Agent Inbox.

Run in production (LangGraph Cloud)

These instructions will go over how to run EAIA in LangGraph Cloud. You will need a LangSmith Plus account to be able to access LangGraph Cloud

If desired, you can always run EAIA in a self-hosted manner using LangGraph Platform Lite or Enterprise.

Set up EAIA on LangGraph Cloud

1. Make sure you have a LangSmith Plus account
2. Navigate to the deployments page in LangSmith
3. Click New Deployment
4. Connect it to your GitHub repo containing this code.
5. Give it a name like Executive-AI-Assistant
6. Add the following environment variables
   1. OPENAI_API_KEY
   2. ANTHROPIC_API_KEY
   3. GMAIL_SECRET - This is the value in eaia/.secrets/secrets.json
   4. GMAIL_TOKEN - This is the value in eaia/.secrets/token.json
7. Click Submit and watch your EAIA deploy

Ingest manually

Let's now kick off a manual ingest job to ingest some emails and run them through our LangGraph Cloud EAIA.

First, get your LANGGRAPH_CLOUD_URL

To kick off an ingest job, run:

python scripts/run_ingest.py --minutes-since 120 --rerun 1 --early 0 --prod ${LANGGRAPH-CLOUD-URL}

This will ingest all emails in the last 120 minutes (--minutes-since). It will NOT break early if it sees an email it already saw (--early 0) and it will rerun ones it has seen before (--rerun 1). It will run against the prod instance we have running (--prod ${LANGGRAPH-CLOUD-URL})

Set up Agent Inbox with LangGraph Cloud EAIA

After we have deployed it, we can interract with any results.

1. Go to Agent Inbox
2. Connect this to your locally running EAIA agent:
   1. Click into Settings
   2. Click Add Inbox
      1. Set Assistant/Graph ID to main
      2. Set Deployment URL to your deployment URL
      3. Give it a name like Prod EAIA
      4. Press Submit

Set up cron job

You probably don't want to manually run ingest all the time. Using LangGraph Platform, you can easily set up a cron job that runs on some schedule to check for new emails. You can set this up with:

python scripts/setup_cron.py --url ${LANGGRAPH-CLOUD-URL}

Advanced Options

If you want to control more of EAIA besides what the configuration allows, you can modify parts of the code base.

Reflection Logic To control the prompts used for reflection (e.g. to populate memory) you can edit eaia/reflection_graphs.py

Triage Logic To control the logic used for triaging emails you can edit eaia/main/triage.py

Calendar Logic To control the logic used for looking at available times on the calendar you can edit eaia/main/find_meeting_time.py

Tone & Style Logic To control the logic used for the tone and style of emails you can edit eaia/main/rewrite.py

Email Draft Logic To control the logic used for drafting emails you can edit eaia/main/draft_response.py