NOTE

This project is currently archived. The relevant API:s and libraries are constantly shifting, and I don't have time to keep up.

signal-aichat is an AI-powered chatbot for the Signal messenger app. It currently supports:

• Bing Chat – !bing
• ChatGPT – !gpt
• Google Bard (untested) – !bard
• HuggingChat (does not support continuous conversations) – !hugchat
• Any local LLM that works with llama.cpp (Vicuna, Alpaca, Koala, et al.) – !llama

Setup instructions

0. Clone this repo

git clone https://github.com/cycneuramus/signal-aichat

1. signald

Chatting with your own number via "Note to Self" does not work. This assumes you will be using a separate Signal account for the bot.

Start the signald container:

docker compose up -d signald

Drop into the container's shell:

docker exec -it signal-aichat-signald /bin/bash

Once inside the container, either:

# link to an existing account:
$ signaldctl account link --device-name signal-aichat

or

# register a new account:
$ signaldctl account register [phone number]

For more information, see the documentation for signald.

Once the account is setup, populate the SIGNAL_PHONE_NUMBER variable in the .env file.

2. Google Bard

See the Bard repository. TL;DR:

Go to https://bard.google.com/

• F12 for console
• Session: Go to Application -> Cookies -> __Secure-1PSID. Copy the value of that cookie.
• In .env, populate the BARD_TOKEN variable with the cookie value

3. Bing Chat

See the EdgeGPT repository. TL;DR:

Make sure to add the exported JSON to the config/bing.json file in this repo directory.

4. ChatGPT

• In the .env file, populate the OPENAI_API_KEY variable with your API key
• Optionally, populate the OPENAI_API_BASE variable to use a different endpoint (defaults to https://api.openai.com/v1)

5. HuggingChat

See the HuggingChat API repository. TL;DR:

• Install the Cookie-Editor extension for Chrome or Firefox
• Go to HuggingChat and login
• Open the extension
• Click Export on the bottom right, then Export as JSON (this saves your cookies to the clipboard)

Make sure to add the exported JSON to the config/hugchat.json file in this repo directory.

6. Llama models

• Place your model weights in the models directory
• In the .env file, change the model in the MODEL path variable to match your model file

7. Additional configuration

In .env:

• Models can be disabled by populating the DISABLED_MODELS variable
• To chat with a default model without explicitly having to trigger a bot response with !<model>, populate the DEFAULT_MODEL variable

Assuming DEFAULT_MODEL=gpt, for example, you'd be able to chat normally:

8. Deploy

• docker compose up -d

And start chatting.