ChatGPT-Telegram-Workers

ChatGPT-Telegram-Workers runs a Telegram AI bot on one Cloudflare Worker with a single KV namespace. Model providers, prompts, permissions, plugins and custom commands are managed in an admin panel instead of environment variables.

Message the bot privately or add it to a group, and it streams answers from the model you configured, keeps history for each chat and can generate images. It speaks four API formats: OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and Cloudflare Workers AI. Any OpenAI-compatible endpoint, such as DeepSeek, Groq or Mistral, works as well. Since version 2.0.0 (September 2026), almost all configuration lives in the admin panel. Send /admin and the panel opens in Telegram as a Mini App, authorized against your Telegram user ID; an optional password also opens it in a normal browser. There you add chat and image providers, each with a base URL, API key, API format and model list, and set the system prompt and history limits. The same panel holds request-template plugins and custom commands that switch the default provider or model. A KV namespace stores chat history, caches and the global config JSON. The shipped wrangler.jsonc declares a Workers AI binding, so Workers AI models run without an account ID or API token. The code also builds for Vercel, Docker and a local Node server. The Deploy to Cloudflare button clones the repository, creates the KV namespace and asks for TELEGRAM_TOKEN and ADMIN_ID, with ADMIN_PASSWORD as an option.

Built for: Individuals and small groups who want their own AI chat bot in Telegram, paid for with their own model API keys

What you can use it for

• Run a private AI assistant inside Telegram
• Give a Telegram group a shared model, with permissions set in the admin panel
• Switch between OpenAI, Anthropic, DeepSeek and Workers AI models with /models
• Generate images in a chat with /img and an OpenAI or Workers AI image model
• Add request-template commands through the plugin system

Deploy

• One-click · main path

  Branch master

  Deploy to Cloudflare

• CLI

  Branch master

  Before deploying, put the KV namespace ID and a vars block with TELEGRAM_TOKEN and ADMIN_ID into wrangler.jsonc, as doc/en/DEPLOY.md shows. Then open the Worker URL and click Bind Webhook.

  git clone https://github.com/TBXark/ChatGPT-Telegram-Workers.git
  cd ChatGPT-Telegram-Workers
  pnpm install
  pnpm wrangler login
  pnpm wrangler kv namespace create DATABASE
  pnpm run deploy

  Copy

Create a bot with BotFather and find your Telegram user ID. After the button deploy, click Bind Webhook on the Worker URL, then add a model provider in the admin panel.

What it needs to run

• Workers AIAI
• KV namespacesDATABASE

Good to know

Telegram is the only chat channel. Every model except Workers AI needs your own provider API key. In groups, the bot sees @mentions only after privacy mode is disabled in BotFather, and public groups also need it added as an administrator. Version 2.0.0 dropped the old environment-variable configuration, so 1.x deployments have to migrate by following MIGRATION.md. Much of the issue tracker is in Chinese.

Cost on Cloudflare

Where this comes from

• Repositoryhttps://github.com/tbxark/ChatGPT-Telegram-Workers
• App websitehttps://chatgpt-telegram-workers.tbxark.workers.dev
• Documentationhttps://github.com/TBXark/ChatGPT-Telegram-Workers/blob/master/doc/en/DEPLOY.md