Updated: 2026-07-24
There are two ways to connect a neural network to a Telegram bot. The usual one: register the bot, get an API key from a neural-network provider, write the code and host it on your own server (and, in some regions, a proxy to Telegram). The no-server way: a Corpis agent already runs on a neural network — the current brain is DeepSeek — so "connecting a bot" here means pasting a @BotFather token into one field.
Create an agent. Open agent creation and fill in the name, personality, rules and greeting — or describe the bot in words and the AI assistant fills the form. The agent answers on a neural network right in the preview.
Check the answers in the preview. Send the agent a couple of messages on the platform and make sure the tone and answers work for you, before connecting Telegram.
Get a token from @BotFather. In Telegram open @BotFather, send /newbot, set a name and @username. You will get a token like 1234567890:AA… — this is the key to your bot, keep it secret.
Open the "Telegram" tab. In the settings of a saved agent go to the "Telegram" section (an unsaved agent does not have it yet).
Paste the token and connect. Paste the token into the field and click "Connect". The platform checks the bot, picks the update-receiving mode and shows a t.me link to it.
Message the bot in Telegram. Open the bot via its t.me link and write to it — it answers on a neural network in your agent voice. The same bot also works in groups, replying when mentioned.
The phrase "bot on DeepSeek" hides two very different amounts of work. If you build the bot yourself, the neural network is an external API: you get a key from a provider, and all the bot code (receiving messages, dialog history, calling the model, error handling) plus the server for it are on you.
Corpis gives you a ready-made brain: the agent is already wired to a neural network, history and the system prompt are assembled by the platform, and Telegram is just another channel of the same agent. No coding at any step.
The agent brain is provider-agnostic: the platform talks to the neural network through a compatible interface, and the specific model is chosen in the agent settings. Right now DeepSeek runs as the brain.
The market has different classes of models — DeepSeek, the ChatGPT family, Claude and others; for the agent they are interchangeable "engines". The point is that the model is a swappable part, not something hard-wired into the bot code: personality, rules, knowledge base and body are stored separately from the brain, so switching the model does not rewrite the agent.
In Corpis the neural network is already connected to the agent. You create the agent with a form (name, personality, rules, greeting) or describe it in words and the AI assistant fills the fields. The finished agent answers on a neural network right away, without a line of code.
To make that same agent speak in Telegram, you paste a @BotFather token on the "Telegram" tab. The platform keeps the connection to Telegram, picks the update-receiving mode and (in blocked regions) reaches Telegram through a proxy. The token is stored on the server and never handed back out.
The connected bot answers exactly like the agent in chat: the same system prompt, the same memory, the same knowledge base and the same brain. Edit the personality or rules and it changes everywhere — in chat, in Telegram, in the site widget and on the agent public page.
So "a bot on a neural network" in Corpis is not a separate project but one of your agent channels. If the model under the hood changes, the bot keeps working with no edits on your side.
Right now the agent brain runs on DeepSeek. The brain is provider-agnostic: the platform talks to the neural network through a compatible interface, and the model is chosen in the agent settings, so switching models does not require rewriting the bot.
No. The neural network is already connected on the Corpis side — you do not get an API key or run a server. All you need is a bot token from @BotFather; no code is written at any step.
You can't pick ChatGPT or Claude yourself — the platform decides the model, and today that's DeepSeek. The brain model is set in the agent settings. ChatGPT and Claude are separate classes of neural networks on the market; which models are available for the agent is decided by the platform, not by your bot code. The agent personality and settings stay the same when the model changes.
Yes. Direct access to the Telegram API is blocked in some regions, so a self-built bot would need a proxy there. In Corpis the platform reaches Telegram through a proxy itself — nothing to configure on the owner side.