Telegram bot powered by OpenAI

Usage

Chat with gpt-4o	Analyze image	Generate with dall-e-3

Roadmap

• Integrate with Assistants API
• Use GPT-4o for conversation
• Use DALL-E 3 for image generation
• Use whisper-1 for speech to text transcription
• Use tts-1 for text to speech synthesis
• Add support for memory recall
• Allow bot to search for old messages, and maybe continue previous conversation
• Add support for reminders
• Add support for Google Mail

Development

# this project uses PNPM instead of NPM
npm install --global pnpm

# install dependencies
pnpm install

# set secrets for the local stage, use `--stage prod` for production
# see the full list of secrets below
pnpm sst secrets set KEY VALUE

# this project use SST to deploy to AWS
# use its dev command for https://docs.sst.dev/live-lambda-development
pnpm dev

Secrets

KEY	VALUE
OPENAI_API_KEY	Create new secret key at https://platform.openai.com/api-keys
OPENAI_ASSISTANT_ID	Create new assistant at https://platform.openai.com/assistants
TELEGRAM_ADMIN_IDS	Comma separated list of Telegram user IDs
TELEGRAM_BOT_TOKEN	Chat with https://t.me/BotFather to create a new bot
TELEGRAM_WEBHOOK_SECRET_TOKEN	Use random.org or similar tool to generate a secret token

• OpenAI assistant doesn't need to be fully configured, the instruction and tools will be set per run
• Only admin can initiate chat with the bot
• Admin can invite the bot to a group chat, it will reply to everyone

Deployment

# set secrets for the production stage
pnpm sst secrets set --stage prod KEY VALUE

# go production!
pnpm run deploy --stage prod

This project has been setup to be deployed automatically with GitHub Actions. A few manual steps are required:

• Setup OpenID Connect to authenticate with AWS https://docs.sst.dev/going-to-production#manual-setup
• Update .github/workflows/sst.yml to use the correct role-to-assume and aws-region