GPT inside your botTelegram bot with ChatGPT:
Telegram bot with ChatGPT:
GPT models without your own API key
A Telegram bot with ChatGPT answers customers with GPT models instead of pre-written button branches. In CHATBOSS.PRO the models are connected over the OpenAI API by the platform, so you need neither your own key nor a server: in the AI block you choose the model — from the cheap GPT-5 Nano to GPT-5.6 Sol of the current generation — and the platform counts the tokens and charges your account balance. CHATBOSS.PRO is an independent bot builder and is not affiliated with OpenAI; ChatGPT remains their product and trademark.
The models are connected over the API on our side: no OpenAI account, no server, no separate dollar invoice. You pick the model per assistant and see what every request costs. From $9/mo.
Author: Павел Седов, founder of CHATBOSS.PRO · Updated: September 19, 2026
Your own OpenAI key
not needed
GPT models
9
Cost per request
in the log
Plans
from $9/mo
What makes wiring ChatGPT into a bot harder than it looks
Calling the API is the easy part. The work is everything around it: hosting, keys, retries, token accounting and keeping the model from saying things you never promised.
Two vendors, two invoices
A self-made bot means a provider account and a card of its own, plus a server bill. Through a builder there is one balance, and the AI cost sits next to the plan.
Spend you only learn about later
Without token accounting a long prompt is discovered at the end of the month. Here each call is logged with the model, the tokens and the price, and the bot metrics show the total for a period.
The model answers about nobody
A raw API call knows nothing about your prices or rules. Getting answers that are actually about your company takes an assistant instruction and a knowledge base — that is what the builder adds on top of the API.
What GPT gives a business bot
GPT here is not a chat for the sake of chatting: it is a step of the scenario that answers, routes, extracts data into a table and hands control to the next block.
Models over the API — no key, no server
Requests to OpenAI models go through the CHATBOSS.PRO infrastructure. You do not open a provider account and you do not maintain a runtime. CHATBOSS.PRO is an independent service and not an OpenAI partner.
The right model for the job
GPT-5.6 Sol for complex consulting, GPT-5.6 Luna as the fast compromise, GPT-5 Nano for a stream of short answers. The model is changed in the assistant settings — the scenario stays as it is.
Your rules of answering
The assistant instruction sets the role, the tone and the boundaries: what to talk about, what to refuse, when to call a human. Up to 50 000 characters of instruction on the senior models.
Token and cost accounting
Every request is written to the AI spend log: model, input and output tokens, cost and the balance after it. Spend per bot is visible in the metrics section.
Answers that appear as they are written
Long replies can be streamed so the text types out in Telegram instead of landing as one message after a pause — the customer sees progress rather than silence.
Boundaries you can enforce
What the bot must not discuss is set in the instruction, and anything outside the topic can be routed to a human branch instead of being improvised by the model.
What a GPT bot is actually used for
The model earns its cost where an answer cannot be laid out on buttons in advance: a free-form question, a comparison, a choice, or a short piece of text written on the spot.
Consulting before the purchase
The bot unpacks the question, compares options from your price list and suggests a fit — then moves the customer on to a form or a payment.
Free-form questions
The customer writes however they like instead of picking a menu item, and GPT answers from the company materials rather than from general knowledge.
Triage and routing
A logical assistant recognises the topic of a message and sends the conversation to the right branch: sales, support, refunds, human operator.
Short generated texts
A product description, a reply to a review, a summary of what the conversation collected — written by the model and passed straight into the next block.
How to connect ChatGPT to a Telegram bot in one evening
The order is the same as for any bot: a token from @BotFather, a scenario in the editor, and an AI block with the model you picked. API access and infrastructure are on the platform.
Step 1
Create the bot and paste the token
Take the token from @BotFather and add the bot in the dashboard — no server, no hosting, no environment to prepare.
Step 2
Add an AI block and choose the model
In the assistant settings pick the model: a senior one for complex conversations, a light one for a stream of similar questions.
Step 3
Describe the rules and the materials
Write the instruction — who the bot is and how it answers — and put your prices, services and FAQ into the assistant knowledge base.
Step 4
Check quality against cost
Run real questions through the test chat, look at the spend log and switch the model until quality and price sit where you want them.
Telegram bot with ChatGPT — frequently asked questions
Is this an official ChatGPT bot?
No. CHATBOSS.PRO is an independent Telegram bot builder that calls GPT models over the API. We are not affiliated with OpenAI; ChatGPT is their product and trademark.
Do I need an OpenAI API key or account?
No. Requests go through the platform infrastructure, so there is no separate provider account to open and no second card to attach — AI is charged to your CHATBOSS.PRO balance.
Which GPT model will answer my customers?
The one you select on the assistant: the current GPT-5.6 generation (Luna, Terra, Sol) or the earlier GPT-5.2, GPT-5 Mini, GPT-5 Nano and GPT-4.1 models. Assistants inside one bot can use different models.
How are GPT requests priced?
By tokens, so the cost depends on the model and on how much text goes in and out. The amount is taken from your balance, every call lands in the spend log, and the total for a period is shown in the bot metrics.
Can I stop the bot from answering off-topic questions?
Yes. The boundaries live in the assistant instruction, and the scenario can be built so that anything outside the topic is handed to a human branch rather than answered by the model.
Will the bot work for customers outside my country?
Yes — Telegram is the only channel involved, so the bot answers wherever your customers are. Plans and AI are billed in US dollars, and the interface is available in English.
Can I keep the conversation history out of the model?
Memory depth is a setting: anywhere from 2 to 100 previous messages, 20 by default, and it can be reduced for sensitive flows. On top of that, end-to-end encryption of conversations can be enabled for the bot.