Telence - Telegram LLM Bot

Telence is a Telegram bot that integrates with large language models (LLMs) to provide contextual, AI-powered responses in group and private chats. It supports multiple LLM providers (OpenAI and Gemini) and leverages conversation history stored in an SQLite database to generate more informed and context-aware answers.

Why Telence?

Telence is designed for seamless Telegram AI interaction, allowing users to chat naturally with a bot that can store conversation history and generate concise summaries. It supports multiple LLM providers (OpenAI, Gemini, and more planned), making it flexible and future-proof.

Built with Deno, it offers a lightweight, secure, and efficient alternative to other chatbot frameworks.

Features

• Multi-LLM Support:
  Easily switch between OpenAI and Google Gemini models using environment variables.
• Context-Aware Responses:
  Stores conversation history in an SQLite database. Retrieves both user and bot messages in chronological order to provide context for generating replies.
• Selective Group Chat Responses:
  In group chats, the bot responds only when explicitly mentioned (e.g., @pjer_bot).
• Commands:
  • /start – Greets the user.
  • /reset – Clears the conversation history (memory reset) for the current chat.
  • /summary <param> – Generates a summary of recent messages. The parameter can either be a number (e.g., /summary 20 for the last 20 messages) or a time period (e.g., /summary 1h for messages from the last hour).
• Logging & Debugging:
  Logs errors and API interactions to bot_logs.txt for troubleshooting.

Technologies Used

• Deno:
  A modern, secure runtime for JavaScript and TypeScript.
• grammY:
  A powerful Telegram Bot framework for Deno.
• SQLite:
  Used for storing chat history and metadata.
• LLM Providers:
  Integrates with OpenAI and Google Gemini for generating AI-powered responses.

Setting Up Your Telegram Bot

To create a new bot, follow the Telegram BotFather guide to generate an API key.

Setup & Installation

1. Clone the Repository:

   git clone https://github.com/andrijeski/telence.git
   cd telence
   

2. Configure Environment Variables:

Create a .env file and define the following variables:

TELEGRAM_BOT_TOKEN=your_telegram_bot_token
OPENAI_API_KEY=your_openai_api_key
GEMINI_API_KEY=your_gemini_api_key
LLM_PROVIDER=openai    # Supported providers: openai, gemini (more planned)
MODEL_NAME=chatgpt-4o-latest
CONTEXT_SIZE=20        # Adjustable context size (number of messages)
BOT_NAME=pjer_bot      # The name of the bot
RELATIVE_TIME_THRESHOLD_SECONDS=600   # Threshold for the system to add time indication in between messages, so the bot is aware when significant time has passed between the chat messages.

3. Create Data Directory:

The bot uses an SQLite database stored in the data directory. Create it before the first run:

mkdir data

4. Run the bot:

Use Deno to run the bot with the required permissions:

deno task start

(The bot's entry point is located at src/main.ts)

Usage

• Start a Chat:
  Send /start to get a greeting and basic instructions.
• Interacting in Group Chats:
  In a group chat, mention the bot (e.g., @pjer_bot) for it to respond.
• Resetting Conversation Memory:
  Use the /reset command to clear all stored messages for the current chat.
• Summarizing Conversation:
  • By Number of Messages:
    /summary 20 will summarize the last 20 messages.
  • By Time Period:
    /summary 1h will summarize all messages from the past hour.

Development

• TypeScript Configuration:
  The project uses a tsconfig.json file with modern settings:

  {
    "compilerOptions": {
      "lib": ["ESNext", "DOM"],
      "module": "ESNext",
      "target": "ESNext",
      "strict": true,
      "noImplicitReturns": true,
      "noUnusedLocals": true,
      "noUnusedParameters": true,
      "allowJs": false,
      "noFallthroughCasesInSwitch": true
    }
  }
  

• Package Management: Basic project metadata, scripts, and import maps are managed via deno.json. Deno automatically handles dependencies, but if needed, deno.lock can be used to ensure version consistency.

Planned Improvements

• Grounding: Grounding or backing up some information using tools like google search. This is crucial for providing up to date information. See feature/grounding branch where a simple form of grounding is already implemented for Gemini 2.5 models.
• Memory Optimization: Summarizing older chat history to keep token costs low while maintaining context.
• Vector Database Support: Implementing memory recall using Pinecone or similar for long-term knowledge retention.
• More LLM Providers: Expanding support beyond OpenAI and Gemini (Mistral, DeepSeek, Claude, etc..).
• Database Security Enhancements: Implement encryption for chat history, restrict database access, and introduce an automated retention policy.

Frameworks & APIs

• grammY (Telegram Bot Framework)
• Telegram Bot API Documentation
• OpenAI API
• Google Gemini API

More Resources

• Deno Documentation
• grammY Telegram Bot Framework
• SQLite for Deno