For a long-polling Telegram bot, choose an always-on worker with automatic restarts and readable logs. For a webhook bot, choose a host with a stable HTTPS URL and fast request handling. That delivery choice matters more than the provider logo.
Most small bots use long polling because it needs no inbound URL. The process calls getUpdates, waits, and calls again. Telegram webhooks reverse the direction: Telegram sends each update to your HTTPS endpoint.
Pick the delivery model first
| YOUR BOT | CHOOSE | THE HOST MUST PROVIDE |
|---|---|---|
| Tutorial bot already using run_polling or bot.launch | Long polling | A process that never sleeps and outbound HTTPS |
| HTTP app with a public route | Webhook | Stable HTTPS, quick responses, and secret verification |
| Bot with slow jobs | Either, plus a worker | A queue or durable job store outside the request |
| Bot behind a private network | Long polling | Outbound access only; no inbound port |
Do not change a working bot from polling to webhooks only to satisfy a hosting platform. The Telegram Bot API supports both. It does not allow getUpdates while a webhook is set, so a half-finished switch simply stops delivery.
A practical provider comparison
| OPTION | GOOD FIT | BILLING AND OPERATIONAL SHAPE |
|---|---|---|
| Spocket | Small Node or Python polling bots, or simple webhooks | $3 fixed monthly price for one app; managed restarts and no root access |
| Railway | A bot plus databases or several connected services | $5 monthly minimum with included usage, then resource-based billing |
| Render worker | A continuous polling process with a Git-based workflow | Background workers are paid services and run continuously |
| Cloud Run | Webhook bots with bursty traffic | Request-based service can scale to zero; always-on CPU needs different settings |
| VPS | Custom system packages, several processes, or full control | Fixed server bill, plus your own patching, TLS, monitoring, and restarts |
These options are not interchangeable. Railway gives broad platform primitives and usage metering. Render has an explicit background worker service. Cloud Run is especially natural for request-shaped webhook traffic. A VPS is the most flexible and the least managed.
What to verify before paying
- The runtime and version your library expects, including Python or Node.js.
- Whether idle services sleep, and what activity the host uses to decide that a service is idle.
- What restarts the bot after an uncaught exception or machine failure.
- Where live and deploy logs are kept, and how long they remain available.
- Whether environment variables can be read back after storage.
- Whether local disk survives restarts. Bot state belongs in a database when it matters.
- How memory, CPU, network traffic, databases, and extra services affect the bill.
Match the host to the bot, not the demo
A command bot that stores little state can run in a small process. A bot that launches Chromium, loads a local model, or processes video needs more memory and may need custom packages. A bot that handles payments needs durable storage, a recovery plan, and monitoring regardless of how little CPU it uses.
A safe migration from your laptop or another host
- 01Deploy without the tokenConfirm dependencies install and the correct entry point starts. An expected missing-token error is easier to read than two bots competing.
- 02Stop the old pollerOnly one process should call getUpdates for a token. Shut down the laptop terminal, old service, and any extra replica.
- 03Set the token on the new hostUse an environment variable rather than committing a .env file. Start one instance and watch the first connection in the logs.
- 04Send a real updateTest one command, one callback button if you use them, and one failure path. Confirm state is written somewhere durable.
- 05Remove the old deploymentOnce the new process is stable, delete or disable the old service so an automatic restart cannot create a polling conflict later.
Spocket is a direct fit when the bot is a Node or Python process and you want fixed per-app pricing. The Telegram runtime page at /run/telegram explains the deployment path. Compare memory and slot limits on /pricing, and use /documentation/logs-and-status when the first start does not connect.