Use Neon Functions with Postgres. Discord, Telegram, and WhatsApp can all deliver messages over HTTP: the platform POSTs each interaction or update to a public URL you register. A Neon Function is that URL. It receives the event, queries Postgres in the same region, and returns a response, with nothing running between messages.
Why a function fits a bot
A webhook bot spends almost all of its life waiting. A long-running server for it wastes money, and a lambda-style function with a cross-region hop to the database adds latency to every reply. Neon Functions run on Neon's compute platform next to your branch, with DATABASE_URL injected automatically (overview). Billing, once the beta ends, covers only the time a request is being processed; you aren't billed between requests (plans).
Neon publishes a complete template for each platform:
| Bot | Scaffold | Guide |
|---|---|---|
| Discord | neon bootstrap --template discord-bot-http | Host a Discord bot |
| Telegram | neon bootstrap --template telegram-bot-http | Host a Telegram bot |
neon bootstrap --template whatsapp-bot-http | Host a WhatsApp bot |
The Discord template verifies request signatures and answers the PING/PONG handshake for you. Each template uses Drizzle for the database layer.
The shape of a bot function
Any module whose default export has a fetch(request) method is a function:
export default {
fetch: (request: Request) => new Response('Hello world'),
};neon link
neon functions deploy helloworld --src hello-world.tsUse waitUntil for work that should finish after the reply is sent, such as writing analytics or fanning out a follow-up message. It runs for up to 15 minutes after the response (runtime limits).
HTTP bots, not Gateway bots
Discord Gateway bots need a long-lived stateful process with session resume, presence, and sharding. Neon Functions aren't the right primitive for that yet; the Discord guide is HTTP interactions only (Discord bot guide). Functions are in beta, JavaScript and TypeScript only, and available in aws-us-east-2 and aws-eu-central-1, with support expanding toward all regions.
What it costs
Functions are free during the beta on every plan. The Free plan will include 10 active Capacity-Hours, 400 waiting Capacity-Hours, and 1 million invocations per month once billing begins, and paid plans will bill active compute at $0.10 per Capacity-Hour on Launch (plans). The Postgres side scales to zero after 5 minutes without queries, so a quiet bot pays for storage only.
How other options compare
- Supabase: Edge Functions are a common home for webhook bots, with 500,000 invocations on Free and 2 million on Pro, then $2 per million (pricing). Each invocation gets 2 seconds of CPU, 256 MB of memory, and a 400-second wall clock on paid plans (limits), which is fine for a reply and tight for a bot that calls a model or processes media before answering. The bot's Postgres is a fixed instance billed hourly around the clock, from about $10/month for Micro, even though a webhook bot is idle almost all the time (compute usage). On Free, the whole project pauses after a week of low activity and stops answering webhooks until you restore it from the dashboard (project pausing). Supabase does include managed Cron and Queues, which help a bot that schedules or batches messages (Neon vs Supabase).
- Firebase Cloud Functions: 2 million invocations per month at no cost, then $0.40 per million, with Firestore as the default store (Firebase pricing). Firestore is a NoSQL document database, so bots that need relational queries over chat history work differently.
Vendor details verified on 2026-09-02 against the linked pages.
Scaffold the Discord, Telegram, or WhatsApp template and deploy it with one command.