Blipko is a conversational personal-budget tracker on Telegram. Log what you spend in plain language — by text or voice, in English, Hindi, Hinglish, Malayalam, or Manglish — and Blipko sorts it into a 50/30/20 budget, tells you what's left, and answers questions about your money. No app to install. No forms. Just a chat.
Most months, the salary just… disappears. Traditional budgeting apps ask you to install something, open it, tap through forms, and pick categories — so most people quit by week two.
Blipko removes every bit of that friction. You already have Telegram open. Type chai 30 and it's logged,
categorized, and counted against your budget in under a second. The discipline of budgeting, with the effort
of texting a friend.
You: chai 30
Blipko: ✅ ₹30 → Wants · Food
Wants left this cycle: ₹14,770 / ₹15,000
You: 🎤 (voice note) "auto eighty to office"
Blipko: ✅ ₹80 → Needs · Transport
You: /status
Blipko: 📊 This cycle — Day 15 of 30
🏠 Needs ▓▓▓░░░░░░░ ₹12,500 / ₹25,000 (50%)
🎯 Wants ▓▓▓▓░░░░░░ ₹9,200 / ₹15,000 (61%)
💰 Savings ▓░░░░░░░░░ ₹2,000 / ₹10,000 (20%)
Safe daily spend left: Needs ₹833/day · Wants ₹420/day
You: can I afford a 4000 pair of shoes?
Blipko: You've got ₹5,800 left in Wants with 15 days to go, so ₹4,000 fits —
but it leaves you ~₹120/day. Doable if you go easy on eating out.
Log a spend the way you'd say it — chai 30, auto 80 office, petrol 500 koduthu, netflix 199. The AI
reads informal, code-mixed Indian languages (English / Hindi / Hinglish / Malayalam / Manglish), extracts the
amount, picks the category, and assigns a bucket. Send a voice note and it's transcribed and logged too.
Every spend lands in Needs (50%), Wants (30%), or Savings (20%) — the proven rule, fully
adjustable. Budgets run on a payday-aware cycle (set your payday; day 1 = calendar month), and your
effective income is max(expected salary, income logged this cycle) so salaried and variable-income users
both work.
A short wizard, not a form: enter your take-home income → tap the spending groups you actually have (common ones pre-ticked) → Blipko auto-creates their subcategories with income-based budget suggestions → choose how hard it should nudge you. Three taps and you're tracking.
Seven parent groups (Essentials, Food & Drinks, Transportation, Entertainment & Leisure, Health & Wellness, Miscellaneous, Savings) break down into leaf categories (Rent, Groceries, Eating Out, Fuel…). Each leaf can carry a monthly limit, editable on the web or via chat.
Beyond logging, Blipko answers free-form questions grounded in your real data via a tool-calling agent:
"how much did I spend on food this month?" · "what's my biggest expense?" · "am I spending more than last month?"
Proactive nudges before you blow a bucket — but opt-in by intensity, default off:
| Dosage | What you get |
|---|---|
| Off | Nothing (default) |
| Gentle | 80% and over-budget warnings, once per cycle |
| Aggressive | + a 50% heads-up and a daily check-in |
| Relentless | Daily check-in and repeated over-budget alerts |
rent 8000 on 1st every month sets up a rule that auto-posts each cycle. Salary, subscriptions, EMIs — set
once, tracked forever.
Sign in with Google at blipko.lol for the full picture: animated overview KPIs, income-vs-spending and per-bucket trends, category breakdowns, a filterable transaction list (expenses + income tabs, CSV export), category & per-category-limit editing, recurring management, and all your settings.
| Command | What it does |
|---|---|
/start |
Set up (or redo) your budget — runs the onboarding wizard |
/status |
Budget health this cycle + safe daily spend |
/report |
This cycle's summary and your biggest leaks |
/recurring |
Manage repeating income/expenses |
/settings |
Change your reminder intensity |
/help |
Full guide to everything the bot can do |
undo |
Remove your last entry |
| (just ask) | "how much on groceries?", "can I afford X?", "biggest expense?" |
| Layer | Tech |
|---|---|
| Backend | Node.js + Express, TypeScript (strict, Clean Architecture) |
| Database | PostgreSQL via Prisma (hosted on Railway) |
| AI parsing | OpenAI gpt-4o-mini (primary) → Gemini 2.0 Flash (fallback), structured output |
| Q&A agent | OpenAI function-calling over read-only data tools |
| Voice | Sarvam AI transcription |
| Messaging | Telegram Bot API (webhook) |
| Scheduler | node-cron (recurring auto-post, nudges, pruning) |
| Web | Next.js 16 (App Router) + React 19, NextAuth v5 (Google), Server Actions |
| Tests | Vitest (unit) + Playwright (API/E2E) |
Two runtimes share one prisma/schema.prisma. The backend follows Clean Architecture; the web app talks to
the DB directly through Server Actions (no REST layer between them).
blipko/ ← Backend: Express + Prisma (TypeScript, CommonJS)
src/
domain/ ← Entities, repository & service interfaces (no outer imports)
application/ ← Use cases + message processors (the business logic)
data/ ← Prisma repositories, AI parsers, Telegram + Sarvam services
presentation/ ← Express routes + Telegram webhook controller
prisma/ ← Shared schema, migrations, seed
└── web/ ← Frontend: Next.js dashboard (Server Actions)
Message flow
Telegram → TelegramWebhookController → ProcessIncomingMessageUseCase
1. ensureUserExists (+ link Telegram ↔ web account)
2. load recent conversation history
3. pre-parse processors (no AI): onboarding wizard, button callbacks, /status, /report,
/settings, /help, undo
4. AI parse → ParsedData { intent, amount, category, bucket, confidence, … }
5. intent processor handles it: Expense · Income · Recurring · Query (ask-anything) · Fallback
Low-confidence parses ask you to confirm the bucket with inline buttons; every Telegram message ID is recorded so duplicate deliveries are dropped.
- Node.js 20+ and pnpm
- A PostgreSQL database URL
- Telegram bot token (from @BotFather)
- OpenAI and Gemini API keys (Sarvam key for voice notes, optional)
git clone https://github.com/square-story/blipko.git
cd blipko
pnpm installcp .env.example .env
# DATABASE_URL, OPENAI_API_KEY, GEMINI_API_KEY,
# TELEGRAM_BOT_TOKEN, TELEGRAM_WEBHOOK_SECRET, SARVAM_API_KEY, PORTpnpm prisma:migrate # apply migrations
pnpm prisma:generate # generate the client
pnpm db:seed # seed the system category taxonomypnpm dev # backend (ts-node-dev) on $PORT
cd web && pnpm dev # dashboard on :3000ngrok http <PORT>
WEBHOOK_URL=https://<your-ngrok>.ngrok.app pnpm webhook:set
# registers https://<ngrok>/api/webhooks/telegram with your secret tokenThe dashboard needs its own env — copy web/.env.example → web/.env.local and fill in AUTH_SECRET,
AUTH_GOOGLE_ID, AUTH_GOOGLE_SECRET, and DATABASE_URL.
pnpm test:unit # Vitest unit tests (processors, budget math, agents)
pnpm test # Playwright API/E2E suite (auto-starts the server)
pnpm lint # ESLintDeployed on Railway as two services (backend + Next.js web) plus a managed PostgreSQL instance, all on one
schema. The backend's pre-deploy step runs prisma migrate deploy; the web build syncs the shared schema
before generating its Prisma client. Live at blipko.lol.
- Weekly habit tracking — weekly budgets and urgency nudges for high-frequency habits (coffee, eating out) to break the "small daily dose" spend pattern.
MIT — see LICENSE. Built by Mohammed Sadik.