← Back to Blog

n8n Telegram AI Agent: Run Your Business From One Chat, With an Approve Button Before Anything Goes Out

A Telegram bot wired to an n8n AI agent lets you log leads by voice note, get a morning brief and approve outgoing texts with one tap. Here is the build, the three settings that keep strangers out, and what it costs as of November 2026.

Tools11 min read•

n8n Telegram AI agent: why a chat beats another dashboard

You are on a ladder when a homeowner calls about a repaint. By the time you are back at the truck you have forgotten half of it, and the follow-up text goes out the next morning, after they have already called someone else. You do not need another app for this. You need somewhere to say "new lead, Dave on Elm Street, two-story exterior, wants a quote next week, text him back" and have it handled.

That is what an n8n Telegram AI agent does well. Telegram is the chat app you already carry, n8n is the automation tool that connects it to your spreadsheet, calendar and texting, and an AI agent in the middle turns loose voice notes into tidy actions. This guide walks through a build a non-developer can finish in an afternoon, with one rule baked in: the agent drafts, you approve, and nothing reaches a customer without your tap.

What operators on this site actually run through Telegram

Telegram shows up as the control panel in six of the documented builds on this site. In 12 n8n Workflows That Replaced $3,000 Per Month in Manual Work, an agency owner runs lead management, onboarding and server alerts through n8n with Telegram and Notion, and reports saving five hours a week on lead management alone. In An OpenClaw Chief of Staff on a $600 Mac Mini That Runs Email, Calendar, and Follow-Ups, a consultant drafts email replies that he approves in about 30 seconds each from his phone, and reports 15 email requests handled the first morning at $50 to $100 a month in API costs. A One-Person Agency Run by 4 OpenClaw Agents on Gemini's Free Tier uses Telegram to direct four agents for about $5 a month in infrastructure. These figures are self-reported by the builders.

The OpenClaw setups are more autonomous and need a machine running all day. The n8n version in this guide is narrower and easier to control: each action is a node you can see, and the risky ones wait for you. If you want the broader always-on assistant, our AI chief of staff guide for solo consultants covers it. If your bottleneck is the inbox, the Gmail assistant guide is the better start.

What it is good at and what it is not

Good fits are short, repeated jobs you would otherwise do on your phone between tasks: logging a lead, checking today's schedule, drafting a follow-up, finding a customer's last quote, or getting a summary of overnight form fills. Telegram's voice notes make this work on a job site, since you talk instead of type.

Poor fits are anything with long back-and-forth, anything that needs several people in the same thread, and anything involving sensitive customer data. Telegram's normal chats, which are what bots use, are not end-to-end encrypted. Keep health, payment and legal details out of it.

A worked example: a painting contractor's pocket office

Picture a three-person painting company. The owner gets 10 to 15 leads a week from the website, Google and referrals, and tracks them in a Google Sheet with columns for name, phone, address, job type, status and next step. Jobs live in Google Calendar.

At 10:40 the owner sends a 25-second voice note to the bot: "New lead, Dave Smith, 412 Elm, two-story exterior repaint, cedar siding, wants a quote next Tuesday or Wednesday, his number is..." The workflow checks the message came from the owner's Telegram account, downloads the audio, transcribes it, and passes the text to the AI agent. The agent adds a row to the sheet, checks the calendar for open quote slots on Tuesday and Wednesday, and drafts a text: "Hi Dave, this is Sam from Brightline Painting. I can come by for your exterior quote Tuesday at 2 or Wednesday at 9. Which works?"

The draft arrives in Telegram with Approve and Decline buttons. The owner taps Approve, Twilio sends the text, and the sheet row changes to "contacted." Total time on the ladder: 25 seconds of talking and one tap. Every morning at 7, a second workflow sends a brief: today's jobs, quotes waiting for a reply for more than two days, and new leads since yesterday.

One voice note, start to finish 1. Telegram Trigger Voice note arrives 2. If: sender is you? Anyone else: stop here 3. Get File + transcribe About $0.003 a minute 4. AI Agent node (chat model + Simple Memory + tools) Tools: add lead to Google Sheets, read today's Google Calendar, draft a text to the customer. It can draft. It cannot send. 5. Send and Wait for Response Draft shown in Telegram with Approve and Decline buttons 6. On approve: send the text Twilio SMS, then mark the lead as contacted in the sheet Separate workflow: Schedule Trigger at 7 am sends the morning brief to the same chat

Step by step: build it this week

Step 1: Create the bot in BotFather

In Telegram, open a chat with @BotFather, send /newbot, and pick a name and a username that ends in "bot." Telegram's docs note the username cannot be changed later. BotFather gives you a token. Treat it like a password: anyone with it can control your bot. If it leaks, send /token to BotFather to generate a new one. Make a second bot now too, named something like "yourbiz test bot." You will need it in Step 2.

Step 2: Add the Telegram Trigger and know the one-webhook rule

In n8n, create a Telegram credential with the token and add a Telegram Trigger node set to the message update. Telegram only allows one webhook per bot, and n8n's docs explain that switching between the test URL and the production URL overwrites it, so one of them silently stops receiving messages. The cleanest fix is the one n8n recommends: use your second bot for testing and the main bot for production. If you self-host, Telegram requires an HTTPS webhook URL, so set n8n's WEBHOOK_URL to your public HTTPS address. n8n Cloud handles this for you.

Step 3: Lock the bot to your account

Every bot has a public link, and anyone who finds the username can message it. Right after the trigger, add an If node that checks the sender's Telegram user ID equals yours, and stop the workflow for anyone else. To find your ID, send the bot a message during a test run and read the message.from.id field in the trigger's output. Do this before you connect any tools. Without it, a stranger could ask your agent to read your lead sheet.

Step 4: Turn voice notes into text

Add a Switch node that checks whether the message has a voice field. For voice, use the Telegram node's Get File operation with the voice file ID and download turned on, then an OpenAI node to transcribe the audio. OpenAI lists gpt-4o-mini-transcribe at $0.003 a minute as of November 2026. For typed messages, pass the text straight through. Both branches should end with a field called text so the agent sees the same input either way.

Step 5: Give the agent a small set of tools

Add an AI Agent node with a chat model, a Simple Memory node keyed to the Telegram chat ID so it remembers the last few messages, and three tools: Google Sheets to append and look up leads, Google Calendar to read events, and a tool that returns a drafted customer message without sending it. Keep the system prompt specific: "You are the office assistant for Brightline Painting. Log new leads to the sheet with every field you can find. Never invent phone numbers or prices. When asked to contact a customer, write one short text under 300 characters and return it as a draft. If anything is unclear, ask one question." Fewer tools means fewer wrong choices.

Step 6: Put an approval in front of anything that leaves the building

After the agent, use the Telegram node's Send and Wait for Response operation with the Approval response type. n8n shows the draft with Approve and Decline buttons and pauses the workflow until you tap one. Turn on Approve Within Chat so you can tap the buttons in Telegram instead of opening a browser link; n8n's docs say this needs your instance on public HTTPS, which n8n Cloud is. Fill in Restrict Who Can Approve with your user ID, and set Limit Wait Time to a few hours so stale drafts expire. On approve, send the text through Twilio and update the sheet row. On decline, reply "OK, not sent."

Step 7: Add the morning brief

Build a second workflow with a Schedule Trigger at 7 am. Read today's calendar events, read sheet rows where status is "contacted" and the last contact date is more than two days old, and pass both to a chat model with the instruction "Write a brief under 150 words: today's jobs in time order, then follow-ups due." Send the result to your chat ID with the Telegram node. This one workflow is often what makes owners keep using the rest.

Step 8: Get told when it breaks

Create an error workflow with an Error Trigger that sends you a Telegram message with the workflow name and error, and set it as the error workflow in each workflow's settings. Google credentials expire and sheet columns get renamed. You want to hear about that from your bot, not from a lead who never got a reply.

What this costs

As of November 2026: the Telegram Bot API costs nothing to use. n8n Cloud Starter is 20 euros a month billed annually for 2,500 executions, and Pro is 50 euros a month for 10,000. Self-hosting the n8n community edition is free apart from your server. Transcription with gpt-4o-mini-transcribe is $0.003 a minute. OpenAI lists gpt-5-mini at $0.25 per million input tokens and $2.00 per million output tokens.

Our estimate for the painting example: about 30 messages a day plus the brief is roughly 1,000 executions a month, inside Starter. At around 4,000 input tokens and 300 output tokens per message, the model costs under $2 a month, and 10 minutes of voice notes a day adds about $1. Twilio texts are $0.0083 per segment plus carrier fees, and you need A2P 10DLC registration before texting US customers from a normal number. Expect 20 to 30 dollars a month all in, before Twilio volume. These are estimates; check your own usage after the first month.

When not to build this

If your team already lives in Slack or Microsoft Teams, build the same flow there instead of adding a new app. If more than two or three people need to use it, a shared Telegram bot gets messy fast, because memory and approvals are per chat. If you handle patient, financial or legal data, keep it out of Telegram. And if you would not check the bot daily, do not build it. An assistant that drafts follow-ups nobody approves is worse than a sticky note.

Frequently Asked Questions

Can n8n build an AI agent that works in Telegram?▾

Yes. Use the Telegram Trigger node to receive messages, the AI Agent node with a chat model, memory and tools to decide what to do, and the Telegram node to reply. It runs on n8n Cloud or a self-hosted instance with an HTTPS address.

Why does my n8n Telegram trigger only work in test or production?▾

Telegram allows one webhook per bot, so activating one URL overwrites the other. n8n recommends a separate bot for testing, or unpublishing the workflow while you test and publishing it again afterward.

How do I stop other people from using my Telegram bot?▾

Add an If node right after the trigger that checks the sender's user ID matches yours and stops for anyone else. For approval steps, fill in Restrict Who Can Approve with your user ID too.

Can the agent ask me before it sends a message?▾

Yes. The Telegram node's Send and Wait for Response operation shows the draft with Approve and Decline buttons and pauses the workflow until you choose, with an optional time limit.

Can it understand Telegram voice messages?▾

Yes. Download the voice file with the Telegram node's Get File operation, transcribe it with a speech-to-text model such as gpt-4o-mini-transcribe at $0.003 a minute, and pass the text to the agent.

How much does an n8n Telegram AI agent cost to run?▾

For a small business, mostly the n8n plan: Cloud Starter is 20 euros a month billed annually as of November 2026. Model and transcription costs at small-business volume are typically a few dollars a month.

Keep Reading

Explore Related Niches

Related Strategies

E-commerceHome ServicesSaaS Product

An OpenClaw agent given $1,000 in startup capital generated $14,700 in revenue in three weeks by autonomously building and selling digital products.

AdvancedUnder $50/moHigh ($2K-10K/mo)
View Strategy
Dev ToolsMarketing AgenciesInternal Tool

Seven AI agents ran a company on a single VPS for a week. They pivoted the business, shipped an MVP with 158 passing tests, then forgot basic housekeeping.

AdvancedUnder $50/moLow (under $2K/mo)
View Strategy
Lead GenHome ServicesFreelance/Agency

19 specialized AI agents running 24/7 for plumbers, HVAC companies, and law firms, responding to leads in under 4 minutes for $8 per month total.

IntermediateUnder $50/moVery High ($10K+/mo)
View Strategy

Keep Exploring the Directory

Browse more case studies, compare tools, or dig into the full tool stack behind these workflows.

Want more strategies like this?

Get weekly AI agent case studies in your inbox.