Why I Built a Real-Time Stock Engine for $0/Month Using Cloudflare Workers

I wanted a personal assistant for the Nepal Stock Exchange (NEPSE) — price alerts, portfolio P/L, market news, and a technical screener — pushed to me on Telegram. What I didn’t want was a VPS to patch, a database to babysit, or a monthly invoice for something only I use.

So I built NEPSE Alert to run entirely on Cloudflare’s free tier. No origin server. No container. No cron box. The whole thing lives at the edge, and the running cost is $0/month.

The whole stack is Cloudflare

The trick is that Cloudflare gives you every primitive a small backend needs, and the free tier limits are generous enough that a personal project never comes close to them.

Concern	Traditional stack	What I used
Compute	VPS / container running 24/7	Cloudflare Workers (pay-per-request, scales to zero)
Database	Managed Postgres	D1 (SQLite at the edge)
Cache / hot state	Redis	KV
Scheduling	cron on a always-on box	Cron Triggers
Telegram delivery	Long-polling process	grammY over webhooks

There’s no process sitting idle waiting for work. A Worker spins up only when a request arrives or a cron fires, does its job in a few milliseconds, and disappears.

Why “scale to zero” is the whole point

A personal tool has an ugly cost profile on traditional hosting: it’s idle 99% of the time, but you still pay for the box to exist so it’s ready the other 1%. NEPSE only trades Monday–Friday during market hours, so a dedicated server would spend most of its life burning money to do nothing.

Workers invert that. I pay (in the free tier, nothing) only for the requests I actually serve and the cron ticks that actually fire. Outside market hours the engine costs literally zero because nothing runs.

Splitting hot state from durable state

The two Cloudflare storage primitives map cleanly onto two different jobs:

• D1 (SQLite) holds everything durable and relational — accumulated daily OHLC history, my holdings, and alert rules. It’s the source of truth, and the technical screener reads weeks of history out of it.
• KV holds the hot, disposable state — the latest price for each symbol and a dedupe key so the same alert doesn’t fire twice. KV is eventually consistent and blazing fast to read at the edge, which is exactly right for “what’s the last price I saw?”

Everything that touches the database goes through a single repository layer (src/db/), so the rest of the code never writes raw SQL. If I ever outgrow D1, that’s the only directory that changes.

// One Worker entry handles HTTP routes AND scheduled events.
export default {
  async fetch(request, env, ctx) {
    // dashboard, JSON API, and the Telegram webhook
    return router.handle(request, env, ctx);
  },
  async scheduled(event, env, ctx) {
    // cron dispatch: price polling, EOD persistence, news digest
    ctx.waitUntil(runCron(event.cron, env));
  },
};

That single index.ts is the entire deployment. wrangler deploy ships it worldwide in a few seconds.

What you actually get

• Price & threshold alerts pushed to Telegram (buy/sell levels, % moves, volume spikes)
• Portfolio tracking with live value and P/L
• A technical screener computing SMA / EMA / RSI / MACD over accumulated history
• A web dashboard with a market overview and a searchable stock picker

All of it on infrastructure that costs nothing when the market is closed and pennies (well, free-tier pennies) when it isn’t.

Try it / read the code

The live dashboard is at stock.bishalstha.info.np, and the full source — 100% TypeScript — is on GitHub.

In the next two posts I’ll go deeper on the hard parts: staying under NEPSE’s rate limits with edge cron, and wiring up the Telegram notification layer inside a stateless Worker.