Calorie Tracker Bot

Because "I'll just estimate the portion size" is how every diet actually goes off the rails.

A Telegram bot that tracks calories and macros (protein/fat/carbs). Just tell it what you ate — in free text or a photo — and it figures out the numbers for you via the Gemini API.

Each user connects their own Gemini API key (stored encrypted) — food recognition and target calculations run under their identity and are billed to their own Gemini balance, not a shared key.

How it works

1. On /start, you pick a language, paste your Gemini API key (validated with a cheap models.list call, then stored encrypted with Fernet), and answer a few questions about yourself.
2. Gemini computes your TDEE and daily calorie/macro targets from that; you can accept them as-is or tweak any single value before they're saved.
3. From then on, just message the bot what you ate — plain text or a photo of the food/label. The message goes to Gemini as your own key, with a prompt asking for strict JSON back: items, grams, and calories/macros per item. You confirm or ask it to redo the read before it's saved.

4. Everything (profile, food log, weigh-ins) lives in a single SQLite file. A background scheduler (APScheduler) pings you once a week to log your weight, and a small per-user "known foods" table lets the bot skip re-guessing a product's macros once it's seen it before.
5. The bot process itself only ever sees your encrypted key — it's decrypted in memory just long enough to make the Gemini call, using your own quota, not a shared one.

Features

• Onboarding that calculates your daily calorie/macro targets from weight, height, age, activity level and goal (AI-proposed, editable field by field)
• Log food by free text or photo — items and macros parsed automatically, with a way to nudge the numbers by hand if the AI's estimate is off, instead of only retyping the whole thing
• Remembers your frequently-eaten foods so it doesn't have to guess twice
• "Remaining today" and per-day history, with an undo/delete option for logged entries
• Weekly weigh-in reminder
• Interface in English, Russian, and Hebrew
• /stats — a private command (owner-only, via ADMIN_TELEGRAM_IDS) showing user counts and daily/weekly activity

Quick start

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

cp .env.example .env
# fill in TELEGRAM_BOT_TOKEN
python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
# paste the result into ENCRYPTION_KEY

python -m bot.main

Deploying on a Raspberry Pi

Copy the project to the Pi (e.g. /home/pi/calorie_bot), fill in TELEGRAM_BOT_TOKEN if you already made a .env (otherwise the installer creates one for you), then run:

./deploy/install.sh

That one script sets up the venv, installs dependencies, generates .env/ENCRYPTION_KEY if missing, installs pm2 if it isn't already on the Pi, starts the bot under it, and enables auto-start on reboot. Auto-restart on crash is just pm2's default behavior — nothing extra to configure. It's also safe to re-run any time (e.g. after a git pull) — every step only acts if it isn't already done.

pm2 logs calorie-bot     # tail logs
pm2 restart calorie-bot
pm2 status

Prefer to do it by hand, or want systemd instead of pm2? deploy/start.sh is the actual entry point (pm2 start deploy/start.sh --name calorie-bot, or point deploy/calorie_bot.service's ExecStart at it and use systemctl as usual) — it just runs the bot, nothing else. The daily DB backup cron job (deploy/backup_db.sh at 3am, logged to logs/backup.log) is set up once by deploy/install.sh; if you're doing everything by hand instead, add it yourself:

crontab -e
# add:
0 3 * * * /path/to/calorie-tracker-bot/deploy/backup_db.sh >> /path/to/calorie-tracker-bot/logs/backup.log 2>&1

Good to know

• Don't rotate ENCRYPTION_KEY once users have saved their API keys — the old keys become permanently undecryptable, and everyone has to re-enter them via Settings.
• The bot is open to any Telegram user by default — no allowlist, since everyone pays for their own Gemini usage anyway.
• Sure, you could ask to use my hosted instance instead of running your own. I wouldn't recommend it — it's running on a Raspberry Pi, and that thing starts sweating at 3 concurrent users, let alone 1000. Spin up your own; it's five minutes and it won't quietly fall over on you.