GrenTeaBot: A Telegram Smart Chatbot Powered by Local Neural Networks

GrenTeaBot is a Telegram chatbot built on local neural networks, integrating Ollama for offline generation and supporting adjustable reply probabilities. It offers Windows users an out-of-the-box AI conversation experience with a focus on data privacy (all computations done locally without third-party servers). Key features include local inference, offline availability, model choice flexibility, and user-friendly setup.

With the popularization of large language models, users increasingly care about data privacy and operational costs. Local AI deployment (no cloud API dependency) has become a trend, as it protects privacy and works in network-restricted environments. GrenTeaBot is a representative of this trend—designed for Telegram, it uses a fully local neural network architecture via Ollama integration, ensuring all conversation generation happens on the user's Windows device without data sent to third-party servers.

Local Neural Network Architecture

• Full local inference: All model computations are done on the user's device; conversation content never leaves the local environment.
• Offline availability: Works as long as Ollama service is running, even without network.
• Model choice: Users can select models based on hardware and needs.

Adjustable Reply Probability

• High-frequency mode: Responds to almost all messages (ideal for channel interaction).
• Low-frequency mode: Triggers replies only under specific conditions.
• Smart balance: Adjusts participation dynamically based on context.

User-Friendly Experience

• Zero programming required: Download and unzip to run.
• Pre-packaged environment: No tedious setup.
• GUI: Intuitive configuration interface.
• Bilingual support: English and Russian.

System Architecture

• Frontend: Python GUI-based desktop app for user interaction and config management.
• Middle layer: Message routing/scheduling module handling Telegram API and local model communication.
• Inference layer: Calls local models via Ollama interface for response generation.
• Config layer: JSON/INI files storing API keys, model params, user preferences.

Ollama Integration

• Model management (download, load, version control).
• OpenAI-compatible API for simplified development.
• Hardware optimization for consumer devices.
• Loosely coupled design via HTTP interface (independent upgrades).

Telegram Bot API Integration

• Webhook mode for real-time message reception.
• Message parsing (text, sender info, context).
• Formatting model outputs into Telegram messages.
• Permission management for channel admins.

System Requirements

• OS: Windows 10+
• Memory: At least 4GB free (for model loading).
• Storage: 300MB+ available space.
• Network: Internet for Telegram communication and model download.
• Account: Valid Telegram account for Bot creation.

Deployment Steps

1. Download the latest Windows package from GitHub Releases and unzip.
2. Create a bot via BotFather to get API Token, fill into config.
3. Install Ollama (if not present) and pull desired models.
4. Run GrenTeaBot and confirm connections to Telegram/Ollama.
5. Add the bot to target Telegram channels and set admin permissions.

Configuration Tuning

Key configs in config.ini/settings.json:

• telegram_token: Bot API key (must be correct).
• model_name: Ollama model (e.g., llama2, mistral).
• reply_probability: 0.0-1.0 (response chance).
• language: Interface/reply language.
• response_style: Bot personality settings. Restart after changes for effect.

Community Interaction Enhancement

• Atmosphere booster: Initiates topics to keep channels active.
• Q&A assistant: Answers common questions to reduce admin burden.
• Entertainment partner: Engages in casual chats for fun.

Privacy-First Communication

• No conversation data passes through third-party servers.
• No worries about cloud privacy policy changes.
• Full control over model behavior/output.

Development & Testing Platform

• Clear code structure for beginners.
• Covers core chatbot functions (message send/receive, model calls).
• Extensible for custom features.

Current Limitations

• Platform restriction: Only supports Windows (macOS/Linux not available).
• Model dependency: Performance depends on Ollama and local model quality; low-end hardware may have poor experience.
• Function boundaries: Lightweight bot without complex multi-turn dialogue or long-term memory.
• Configuration barrier: Bot creation and Ollama installation may challenge some users.

Potential Improvements

1. Cross-platform support (macOS/Linux).
2. Containerized deployment (Docker images).
3. Web-based config interface (replace local GUI).
4. Plugin system (custom extensions like domain-specific knowledge bases).
5. Dynamic model switching during runtime.

GrenTeaBot demonstrates a feasible path for local AI applications. It proves ordinary users can run practical AI chatbots on personal computers without cloud computing support. This "edge AI" paradigm reduces costs and returns data sovereignty to users. For those wanting Telegram AI assistants with privacy concerns, GrenTeaBot is a worthy solution. With local model tech advances and hardware improvements, similar projects will likely offer better experiences and broader value.