Fogless

A personal expense tracker that runs off plain text. You type what you spent (or received), and it gets parsed, categorized, and logged — no forms, no dropdowns, no required app structure to follow. Built as a Flask serverless backend on Vercel with a Supabase (Postgres) database, an optional Telegram bot, and a small web frontend.

• Natural language parsing — type something like 24 coffee upi or got 424 upi and it extracts the amount, category, payment method, and whether it's income or an expense. Uses Groq (openai/gpt-oss-120b) for parsing, with a local regex-based fallback if the API is unavailable or no key is set.
• Multiple transactions in one message — 120 lunch upi, 50 bus cash, 200 veggies card gets split and logged as three separate entries.
• Income vs. expense detection — words like "got," "received," "salary," "refund," or "credited" are treated as income signals; everything else defaults to expense.
• Voice input — record a short voice memo describing a purchase and it's transcribed via Groq Whisper (whisper-large-v3-turbo) and parsed the same way as typed text.
• Slash commands — /stats, /balance, /history, /undo, /delete, /aliases, /reset for quick ledger operations without needing full sentences.
• Telegram bot — the same parsing and command logic is available through a Telegram webhook, so you can log expenses by messaging the bot directly instead of opening the web app.
• Per-method balances — tracks running balances separately for cash, UPI, card, and netbanking, plus a combined total.
• Works without a database — if Supabase isn't configured, it falls back to in-memory storage, useful for local testing.

You send a message — through the web UI or Telegram — and it's routed through the same parsing pipeline either way. If it starts with /, it's treated as a command. Otherwise it goes to Groq (or the fallback parser) to extract one or more transactions, which get written to Supabase and reflected immediately in your balance totals.

Example:

110 Food upi 100 Juice cash 100 biscuit

gets parsed into three separate transactions, each logged individually.

• Python 3.9+
• A Groq API key (optional — enables LLM parsing and voice transcription; without it, a local rule-based parser is used instead)
• A Supabase project (optional — enables persistent storage; without it, data lives in memory and resets on restart)
• A Telegram bot token (optional — enables the Telegram integration)

git clone https://github.com/ScarsAndSource/fogless.git
cd fogless
pip install -r requirements.txt

Create a .env file in the project root:

BOT_TOKEN=your_telegram_bot_token
OWNER_ID=your_telegram_chat_id
TELEGRAM_SECRET_TOKEN=your_optional_secret_token
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_KEY=your_supabase_service_role_key
GROQ_API_KEY=gsk_your_groq_api_key

All of these are optional — the app degrades gracefully without them, though you'll lose the corresponding feature (LLM parsing, persistence, or the Telegram bot).

python api/index.py

Open http://localhost:5000.