Telegram Summarizer Bot — OpenAI + Gemini + python-telegram-bot + Fly.io

A daily summarizer bot for Telegram group chats. At 23:59 (Europe/Kyiv) it posts #Підсумки_дня (Daily Summary): a list of topics discussed during the previous day with links to the first message of each topic and a link to the initiator (clickable user link or @username).

• Python 3.11+, uv for dependency management
• python-telegram-bot v21 (uses JobQueue for the daily job)
• OpenAI GPT and Google Generative AI (Gemini) to cluster messages into topics
• Configurable toxicity levels (0-9) for different response styles
• Token-aware message processing using tiktoken for optimal API usage
• SQLite (optional Fly.io volume for persistence across releases)
• Runs in long-polling mode

1. Telegram Bot Token from @BotFather
   • Create a bot → /newbot
   • Disable Group Privacy: /setprivacy → pick the bot → Disable (so the bot can read all group messages)
   • Add the bot to your chat (giving it admin rights is recommended)
2. API Keys:
   • OpenAI API Key (for GPT models)
   • Gemini API Key (Google Generative AI)
3. uv and Python 3.11+ installed locally
4. (Production) flyctl account + CLI

If the repo already has pyproject.toml and src/main.py, jump to step 3.

1. Clone and setup

git clone <your-repo> cd xxl-bot-summarizer uv sync

2. Install additional dependencies (if needed)

uv add "python-telegram-bot==21._" "google-generativeai==0._"
"openai>=1.102.0" "tiktoken>=0.11.0" "orjson==3.*"

3. Create .env in the project root (example)

TELEGRAM_BOT_TOKEN=123456789:AA...your_token... OPENAI_API_KEY=sk-...your_openai_key... GEMINI_API_KEY=AIza...your_gemini_key... TZ=Europe/Kyiv
# Configure which chats use which AI provider (comma-separated chat IDs):
OPENAI_CHAT_IDS=-1001234567890,-1009876543210 GEMINI_CHAT_IDS=-1001111111111,-1002222222222
# Optional:
OPENAI_MODEL_NAME=gpt-4o-mini GEMINI_MODEL_NAME=gemini-1.5-flash
# If you mount a volume on Fly (see below):
# DB_PATH=/data/bot.db

4. Run locally

uv run python src/main.py

5. Test in a chat

• In a test group, send several messages and replies
• Run /summary_now [0-9] — you should see today's summary with specified toxicity level
• Run /summary_now — uses maximum toxicity (level 9) by default
• /chatid returns the chat id and shows which AI provider is configured
• Use /enable_summaries and /disable_summaries to control automatic daily summaries

Alternatives to get chat id without /chatid:

• Copy the link to any message in a private supergroup: https://t.me/c/<internal_id>/<msg_id> → your chat_id is -100<internal_id>.
• Use Bot API: curl -s "https://api.telegram.org/bot$TOKEN/getUpdates" | jq '.result[].message.chat' (run deleteWebhook first if needed).

• /summary_now [0-9] — Generate immediate summary with optional toxicity level (0=friendly, 9=maximum toxicity)
• /chatid — Show chat ID and configured AI provider
• /enable_summaries — Enable automatic daily summaries
• /disable_summaries — Disable automatic daily summaries
• /status — Show current configuration status

The bot supports both OpenAI and Gemini models. Configure which chats use which provider via environment variables:

• OPENAI_CHAT_IDS: Comma-separated list of chat IDs that will use OpenAI (GPT) models
• GEMINI_CHAT_IDS: Comma-separated list of chat IDs that will use Gemini models

Each chat can only use one provider. The bot will automatically use the appropriate API based on the chat configuration.

• Level 0: Friendly and positive tone
• Level 1-3: Light humor and mild criticism
• Level 4-6: Pronounced sarcasm and sharp criticism
• Level 7-9: Maximum toxicity and provocative style

# Linux/macOS
curl -L [https://fly.io/install.sh](https://fly.io/install.sh) | sh fly auth login

The project includes a uv-based Dockerfile:

FROM ghcr.io/astral-sh/uv:python3.11-bookworm 
WORKDIR /app
COPY pyproject.toml uv.lock ./ 
RUN uv sync --frozen --no-dev
COPY . .
CMD ["uv", "run", "python", "src/main.py"]

fly launch --no-deploy
# pick a region close to Ukraine/Europe (e.g., fra/ams/waw)

This creates fly.toml (defaults are fine).

By default, the SQLite file inside the container is ephemeral between releases. To persist history:

fly volumes create data --size 1

Add to fly.toml:

[[mounts]]
  source = "data"
  destination = "/data"

And set the DB path as a secret:

fly secrets set DB_PATH=/data/bot.db

Import from .env:

fly secrets import < .env
# or set specific ones
fly secrets set TELEGRAM_BOT_TOKEN=...
OPENAI_API_KEY=...
GEMINI_API_KEY=...
TZ=Europe/Kyiv
OPENAI_CHAT_IDS=-100...
GEMINI_CHAT_IDS=-100...

Updating secrets triggers a rolling restart automatically.

fly deploy

Verify:

fly status
fly logs

• Code or Dockerfile changed: fly deploy
• Secrets/env only: fly secrets import < .env (auto-restart)
• Restart current release without rebuilding: fly apps restart <app-name>

• /chatid in the production chat returns a negative id (-100… for supergroups) and shows the correct AI provider
• produces a friendly summary /summary_now 0
• /summary_now 9 produces a highly toxic summary
• produces a summary with clickable links to the first message and to the initiator /summary_now
• A post arrives at 00:00 Europe/Kyiv (if summaries are enabled) #Підсумки_дня
• BotFather Group Privacy = Disabled
• Both OpenAI and Gemini chats work correctly

• Multiple AI providers: Each chat must be configured for exactly one AI provider (OpenAI or Gemini)
• Token limits: The bot uses tiktoken to efficiently manage token usage and stay within API limits
• Private supergroups: message links look like https://t.me/c/<internal_id>/<msg_id> and work for chat members
• HTML escaping: Names/titles/summaries are automatically escaped to avoid broken markup
• Event loop: Uses JobQueue from PTB to avoid event-loop conflicts
• JSON responses: Both providers are configured for JSON output with appropriate fallbacks
• Safety filters: If high toxicity levels are blocked, the bot automatically retries with lower levels

• This event loop is already running → don't call run_polling() inside asyncio.run(...); use the provided structure src/main.py
• no running event loop with schedulers → the project uses PTB JobQueue (see ) src/scheduler.py
• getUpdates conflict → run deleteWebhook before getUpdates to fetch chat id
• No midnight summary → check TZ, JobQueue schedule, summaries enabled via /enable_summaries, and that privacy is disabled
• API errors → check that the chat is configured in the correct *_CHAT_IDS environment variable
• Safety blocks → try lower toxicity levels or check API quotas/content policies