A Guide for Creating a Telegram Bot using local LLM 

Objective: This guide provides a step-by-step overview of how to build a functional Telegram chatbot that runs on your own computer, using a free, locally hosted Large Language Model (LLM).

Step 1: Create the Telegram Bot

Register the bot with Telegram to get an API key using the BotFather bot.

1. Find BotFather: Search for @BotFather on Telegram.
2. Create Bot: Send the /newbot command and follow the prompts to set a name and username.
3. Get API Token: BotFather will provide a token. This key allows your Python script to control the bot.

Step 2: Set Up the Local AI (Ollama)

Use Ollama to run AI models locally for free. It manages downloading and running the models.

1. Install Ollama: Download and install Ollama for your OS. It runs as a background service.
2. Download a Model: Choose a model based on your needs for speed vs. intelligence and download it via the terminal.

• Smaller Models (e.g., phi, gemma:2b): Faster, use less memory. Good for prioritizing speed.
• Larger Models (e.g., mistral, gemma:7b): Smarter, but much slower without a GPU.

Use this command to download a model:

ollama pull <model_name>

3. Verify Installation: List all local models with this command:

ollama list

The Ollama service runs in the background, making the model available to your script.

Step 3: The Python Script

The Python script connects the Telegram bot to the Ollama AI.

Core Functionality:

1. Bot Personality (System Prompt): A “system prompt” variable defines the bot’s personality and rules. The AI uses this to guide all its responses.

# Example system prompt
	system_prompt = """
    		You are a helpful AI assistant.
   		Keep answers friendly and concise.
    		"""

2. Handling Messages: A function is triggered when a user sends a message.

3. Communicating with Ollama: This function sends the user’s message and chat history to the local Ollama service to get a response.

# import ollama
	response = ollama.chat(
	    model="your_chosen_model_name",
	    messages=conversation_history
	)
	ai_response = response['message']['content'].strip()

   4. Implementing Memory (Database): The bot saves chat history to a local SQLite database file for permanent memory. It reads the history before a request and saves the updated history after a response.

Step 4: Run the Bot

Run the main Python script from your terminal to start the bot.

python main.py

The script will poll for new messages, and the bot will be active.

Step 5: Final Considerations

Understanding the trade-offs of running AI locally is key.

• Speed: Running on a CPU instead of a dedicated GPU is the main bottleneck. This causes slow response times.
• Intelligence: There is a direct trade-off between speed and intelligence. Smaller, faster models are less capable and more prone to making mistakes or “hallucinating”.
• Hardware: Models are resource-intensive. They require significant RAM and can max out your CPU, slowing down your entire computer while generating a response.