Where to Host an Always-On Telegram AI Bot Without a Server

AgentRoost · May 17, 2026 · 7 min read · View as Markdown
AgentRoost — Comparisons

You want a Telegram AI bot that answers questions, summarizes articles, monitors something, or just keeps context across days — and you want it running around the clock. The question is how you actually get there.

There are two honest paths: build it yourself on a server, or use a platform that handles the infrastructure. Here's the full picture of both.

The DIY Path: What It Actually Takes

Running a persistent Telegram AI bot on your own means stitching together several moving parts. Here's the real checklist:

1. Get a VPS

You need a machine that stays on 24/7. A cheap VPS (Hetzner, DigitalOcean, Vultr) runs $4–12/month depending on region and specs. That's the floor — it doesn't include anything else.

2. Register the Bot with BotFather

On Telegram, open @BotFather, run /newbot, pick a name and username, and grab your BOT_TOKEN. This is the key Telegram needs to send messages to your bot.

3. Wire Up an AI API Key

Your bot needs to call an LLM somewhere. That means:

  • Creating an account at OpenAI, Anthropic, Google, or a proxy like OpenRouter
  • Adding payment info and setting spending limits
  • Injecting the API key into your bot's environment

This is a separate recurring cost on top of your VPS. If your bot is chatty or handles file summarization, the AI bill can spike unexpectedly.

4. Write and Deploy the Bot

Pick a framework (python-telegram-bot, Telegraf for Node, grammY, etc.), write your polling/webhook handler, and push the code to your server. A minimal webhook-based bot needs something like this in your handler:

async def handle_message(update: Update, context: ContextTypes.DEFAULT_TYPE):
    user_text = update.message.text
    reply = await call_llm(user_text)  # your API call here
    await update.message.reply_text(reply)

Straightforward enough — until you need conversation memory, file handling, or scheduled tasks.

5. Keep It Alive

Your process will crash. You need:

  • systemd or supervisor for process management
  • Log rotation so the disk doesn't fill up
  • Monitoring so you know when it's down
  • A way to update it without breaking the bot mid-conversation

This is the part DIY guides gloss over. In practice it's an afternoon of setup per bot, and occasional weekend debugging when something drifts.

6. Manage Memory (If You Want It)

A basic bot forgets everything between restarts. Persistent memory — knowing that a user asked about topic X last Tuesday — requires storing conversation state in a database, implementing retrieval logic, and deciding how to trim context windows when they get long.

None of this is impossible. But the total DIY cost looks like this:

Component Monthly cost
VPS (2 GB RAM, EU) $6–12
LLM API (moderate usage) $5–20
SSL/nginx/domain $0–5
Your time to maintain hours
Total $11–37+ / month

And that's before you build anything useful.


The Hermes Path: What Changes

Hermes on AgentRoost is a persistent AI assistant framework designed for exactly this use case. Instead of assembling pieces, you get a running assistant in roughly two minutes:

  1. Sign up at agentroost.app — email/password, Google, Microsoft, or Discord
  2. Pick the Hermes framework from the workspace menu
  3. Name your assistant (this becomes your private instance)
  4. Connect Telegram in one click — the AgentRoost manager bot sends you a link; open it, /start your agent's Telegram bot, and it's live

No BotFather. No API keys. No server. The Telegram bot is auto-provisioned.

What Hermes Includes Out of the Box

Persistent memory across days. Hermes remembers context between conversations — no database to wire up, no retrieval logic to build. Ask it to track something on Monday; follow up on Thursday and it knows what you meant.

Scheduled tasks. You can ask Hermes to check something every morning, summarize your inbox at noon, or ping you when a condition is met. The scheduling layer runs on AgentRoost's infrastructure, not a cron job you have to babysit.

AI credits included. This is the part that surprises people: the LLM calls are already paid for. You're not managing an OpenAI account or watching your credit balance. Every plan includes a credit pool for AI usage — no separate API key, no separate bill, no surprise charges when usage spikes.

350+ LLM models, switch anytime. You can change which model powers your assistant without touching a config file.

What It Costs

Plans start at $19.99/month all-in — that's the server, the AI credits, the Telegram provisioning, and the uptime guarantee rolled into one number. Compare plans for Plus and Pro tiers if you need more compute or larger credit pools. Monthly billing, cancel anytime, 14-day money-back guarantee.


Honest Comparison

DIY (VPS + BotFather) Hermes on AgentRoost
Setup time 2–6 hours ~2 minutes
BotFather required Yes No (auto-provisioned)
API key management Yes (separate account + billing) No (credits included)
Persistent memory Manual (you build it) Built-in
Scheduled tasks Manual (cron/systemd) Built-in
Uptime monitoring Manual Handled
LLM model choice Whatever the API supports 350+ models
Monthly cost $11–37+ (variable) From $19.99 (fixed)
What breaks at 2am Your problem Not your problem

Who Should Pick DIY

DIY is still the right call if you need deep control over the bot's code, you're already running a VPS for other reasons, or you want to build something highly custom that no hosted framework would support. A Python developer building a bot with unusual integrations or specific security requirements is probably better off self-hosting.

Who Should Pick Hermes

Hermes makes sense if your goal is the assistant, not the infrastructure. If you want a bot that monitors your email, summarizes research, tracks a project, or answers questions from your team — and you want it working this afternoon rather than this weekend — the hosted path is faster and often cheaper once you count the AI API bill.


What You Can Actually Build With Hermes

To make this concrete: here are assistant patterns that work well with Hermes today:

  • Daily briefings: ask Hermes to summarize a set of topics every morning and message you on Telegram before 9am
  • Research assistant: send it a URL or a block of text; it summarizes, extracts key points, and remembers what you've already read
  • Inbox triage: forward email summaries; Hermes tracks threads and follows up on things you flagged
  • Monitoring alerts: check an API or a URL on a schedule and notify you when something changes
  • Meeting notes: paste transcript text; Hermes extracts action items and stores them for later recall

These are all things you could build on a VPS — but with Hermes you're describing the task, not writing the plumbing.


Get Started

If you want to try the Hermes path: start here. Pick the framework, name your assistant, connect Telegram in one click. The AI credits are already included — no BYOK, no API account, no separate bill.

If you want to compare all frameworks first (including n8n for workflow automation and OpenClaw for a lighter personal assistant): see all agents.

The infrastructure is someone else's problem. The assistant is yours.

Frequently asked questions

Do I need my own OpenAI API key to use Hermes?

No. AI credits are included in every AgentRoost subscription. You don't need to create an OpenAI (or any other LLM provider) account, and there's no separate AI bill. The credits for LLM calls are part of your plan.

Do I need to set up a Telegram bot manually through BotFather?

No. When you launch a Hermes workspace, AgentRoost auto-provisions a private Telegram bot for your instance. You connect it in one click from the AgentRoost manager — no BotFather, no token copying, no webhook configuration.

What happens to my bot if I cancel?

Your Hermes instance is private to your account. If you cancel, the workspace stops. There's a 14-day money-back guarantee if you want to try it and change your mind.

Can I switch which AI model powers my Hermes assistant?

Yes. AgentRoost supports 350+ LLM models and you can switch models at any time from the workspace settings — no config file editing, no restart required.

Is Hermes suitable for a team, or is it just personal?

Hermes is designed around a single persistent assistant per workspace. It works well for individual use or a small team where one person runs the assistant and shares its Telegram bot. For team-wide automation with branching workflows, the n8n framework on AgentRoost is a better fit.