/gptemailsummary

Uses Gmail and OpenAI's APIs to summarize emails

Primary LanguagePython

GPT Email Summary

Uses IMAP and OpenAI's APIs to summarize emails through a python server and iOS Shortcut. This usecase was inspired by Justin Alvey's post on Twitter, which I was unable to find the source code for, and decided to code manually through ChatGPT. https://t.co/TLIoW48rLg

SCR-20230416-j4q

This code uses the IMAP Protocol to call for the latest unread emails from your inbox. It uses HTML and code filtering to avoid hitting OpenAI's token limit for marketing based emails. The contents of those emails are given to OpenAI with pre-written instructions (By Justin Alvey with minor additions) to summarize the emails. That output is given to an iOS Shortcut, which is used on MacOS or iOS to start the process and return the result. Siri will read out the summary, or code has been provided in the Shortcut to integrate with ElevenLabs. All emails will also be marked as read after successfully being sent to OpenAI.

This process requires the python script to be running constantly on a home computer or Raspberry Pi. It has been packaged as a Docker container to be able to run automatically on startup. To run this script outside of your home network, it will also require a (free) Cloudflare Tunnel or something similar. See the instructions below.

Requirements before Installation:

Email Provider

  1. Enable IMAP in your account settings.
  2. Gmail users may need to allow Less Secure App Access in their Google Account Settings, or use an application password.

OpenAI

  1. Go to platform.openai.com on your web browser.
  2. Create an account by clicking on the "Sign up" button at the top right corner of the page, then follow the prompts to create your account.
  3. After your account has been created, click on your account profile at the top-right corner of the page and select "API Keys".
  4. Click on "+ Create new secret key" to create a new API key.
  5. Copy the API key that appears on the screen and save it in a safe place. This API key is required to authenticate your Python script when making requests to OpenAI's API.

ElevenLabs

  1. Go to https://beta.elevenlabs.io and click your Profile
  2. Your API Key is visible there.
  3. Go to https://api.elevenlabs.io/docs with your API Key to retrieve your Voice ID using Get Voices
  4. Save your API Key and Voice ID for use in the iOS Shortcut.

iOS Shortcut

  1. The iOS Shortcut is the method used to trigger the email summary from your iPhone or Mac. Frankly I have no clue how to control this from Android or Windows, as my knowledge of Python is non-existant and I used ChatGPT to code this.
  2. Shortcut (ElevenLabs): https://www.icloud.com/shortcuts/820a1be086d444168fa14312230aa093
  3. Shortcut (Siri): https://www.icloud.com/shortcuts/f87f7b4839064904896ce642f068581f
  4. The Shortcut includes setup instructions. Enter the port of the computer running this script, or the Cloudflare Tunnel URL. This app uses port:1337

Cloudflare Tunnel

  1. To access your Python Server outside of your home (or any other self-hosted applications you may have running) I recommend using a Cloudflare tunnel as described by Crosstalk Solutions
  2. https://www.youtube.com/watch?v=ZvIdFs3M5ic
  3. For Home Assistant users, the Cloudflare tunnel can be run as an add-on to your installation
  4. https://www.youtube.com/watch?v=xXAwT9N-7Hw

Script Installation

  1. Download GmailSummary and put it in a folder of your choice
  2. Install Docker: curl -sSL https://get.docker.com | sh
  3. Install Portainer, a GUI interface to manage Docker containers and Environment variables sudo docker run -d -p 9000:9000 --name=portainer --restart=always -v /var/run/docker.sock:/var/run/docker.sock -v portainer_data:/data portainer/portainer-ce
  4. Access Portainer and create login to view Docker containers: http://localhost:9000
  5. Build the Docker Container: Set your terminal directory with cd /home/pi/Documents/GmailSummary, then run sudo docker build -t gmailsummary . to build the Docker Container.
  6. Modify the following command. Place your OpenAI API Key inside the quotation marks of your-api`key. Choose "gpt-4" or "gpt3.5-turbo" and use that as OPENAI_ENGINE parameter. Example: sudo docker run -d --name gmailsummary -p 1337:1337 -e OPENAI_API_KEY="your-api-key" -e OPENAI_ENGINE="gpt-4" gmailsummary
  7. View Container Logs in Portainer: Home > local > Containers > GmailSummary > Quick actions: Logs
  8. Run the Script: https://www.icloud.com/shortcuts/9d5749b5c54d4162a7a47be6f862cb25
  9. Update Environment Variables: Home > local > Containers > GmailSummary > Duplicate/Edit > Env > Deploy the Container

Planned Features

  1. Support for multiple inboxes ✅
  2. WebUI (draft is almost visually complete but data from the python script is not connected to the page)
  3. Dynamically change quantity of emails to fill 90% of OpenAI Token Limit ✅

Usage

Configure the following Environment Variables to change results as needed:

  1. OPENAI_API_KEY = Default: "". Your API Key from OpenAI.
  2. OPENAI_ENGINE = Default: gpt-4. Configure to use gpt-4, gpt3.5-turbo, or a different chat model. This variable should read "Model", will change at some point.
  3. OPENAI_MAX_TOKENS = Default: 1000. Use to set the maximum length of the return message. Drastically raising this may result in a 400 error (max tokens) depending on the size of the payload sent.
  4. OPENAI_TEMERATURE = Default: 0.7. Temperature controls response randomness and creativity.
  5. CUSTOM_PROMPT = Default: Pretend to be a friendly assistant to someone that you know really well. Their name is Daniel, and they have just asked if there are any noteworthy new emails. Respond providing relevant summaries and if there are any important details or followups needed for each of the emails without just reading them out. Maybe slip in a joke if possible. Try to be observant of all the details in the data to come across as observant and emotionally intelligent as you can. Don't ask for a followup or if they need anything else. The emails to summarize are included below. Don't include emojis in your response. Write the response in a way that works best spoken, not written. Don't read out URLs or two-factor authentication codes.. Customize as needed depending on the content of your inbox. Make sure to update the name in the prompt.
  6. EMAIL_MAXCHARACTERS = Default: 25000. Sets the maximum amount of characters for a payload to OpenAI when EMAIL_VARIABLEQUANTITY=true. It was easier than calculating tokens, but less precise. Typically it's a 4:1 ratio of characters to tokens, with an 8,000 token limit between the sent and received tokens. A value of 28000 has resulted in a 400 error (max tokens) before, so I lowered the default.
  7. EMAIL_MAXEMAILS = Default: 10. Sets the maximum amount of emails to use for a payload to OpenAI when EMAIL_VARIABLEQUANTITY=false. This may be useful if you want to ask for more detail about your emails by focusing on a smaller set of data.
  8. EMAIL_VARIABLEQUANTITY = Default: true. Sets the script to use a character limit when true, or a fixed number of emails when false to create the OpenAI payload.