projects/pi-telegram-247/README.md

Pi 24/7 Telegram Agent on Render

Running Pi coding agent as a background service with Telegram interface.

Architecture

Render Free Tier (ephemeral)
├── Pi (headless RPC mode, --no-session)
│   └── pi-telegram-connect extension
├── Telegram Bot (via extension)
└── No persistent storage

Note: Render free tier has no persistent storage. Sessions reset on restart. This is fine for a 24/7 agent - it keeps general context in the system prompt, not session history.

Prerequisites

1. Render Account - https://render.com
2. Telegram Bot Token - From @BotFather
3. Pi API Key - From your LLM provider (Anthropic/OpenAI)
4. GitHub Repo - To deploy from

Quick Deploy

1. Fork/Clone This Repo

git clone https://github.com/yourusername/pi-telegram-247.git
cd pi-telegram-247

2. Set Environment Variables in Render

Variable	Description
TELEGRAM_BOT_TOKEN	Bot token from @BotFather
ANTHROPIC_API_KEY	Or OPENAI_API_KEY
PI_PROVIDER	anthropic or openai
PI_MODEL	e.g., claude-sonnet-4-20250514

3. Deploy to Render

1. Connect GitHub repo
2. Select Background Worker type
3. Use render.yaml for auto-config
4. Deploy

Local Development

# Install dependencies
./scripts/setup.sh

# Configure
cp .env.example .env
# Edit .env with your keys

# Run
./scripts/run.sh

How It Works

Pi runs in RPC mode (headless) with the pi-telegram-connect extension:

1. Pi starts with pi --mode rpc
2. Extension connects to Telegram via long polling
3. Messages flow: Telegram → Extension → Pi → Response → Telegram
4. Session persists in /data/pi-sessions/

Telegram Commands

• /start - Show help
• /new - New session
• /stop - Abort current turn
• /status - Show bot status

SCRAPER TODO

• Channel scraper using Telethon
• AI-ready JSON output
• Resume capability
• Media download support

References

• Pi RPC Mode
• pi-telegram-connect
• Pi Extensions
• Render Background Workers