GPT-53: DNS-based LLM Chat System

A complete system for chatting with Large Language Models (LLMs) through DNS TXT record queries. Includes a TypeScript DNS server and Python client for secure, authenticated AI conversations.

Designed for restricted network environments where only DNS queries are permitted (like unpaid planes WiFi 😉), enabling access to models like OpenAI o1, GPT-4, Claude, and others. This DNS-based communication can be extended to other applications like chat rooms, file transfer, or any data exchange over DNS infrastructure.

Demo

demo.mp4

Usage

DNS Query Format

Predefined Endpoints (no authentication):

• PING → Returns PONG
• LIST → Returns available models

Chat Format: [10-char API key][1-char model index][user prompt]

Direct DNS Queries

# Test server connectivity
dig @127.0.0.1 -p5333 TXT "PING"

# Get available models
dig @127.0.0.1 -p5333 TXT "LIST"

# Chat examples (API key: my-api-key, and using Model 0)
dig @127.0.0.1 -p5333 TXT "my-api-key0What is AI?"
dig @127.0.0.1 -p5333 TXT "my-api-key0Tell me a programming joke"
dig @127.0.0.1 -p5333 TXT "my-api-key0How do I center a div in CSS?"

Python Client Usage

Python client

# Call using python
python dns_chat.py ping

# Or install it for direct use
pip install -e .
gpt53 ping

# Example usage
gpt53 --api-key my-api-key --host 127.0.0.1 --port 5333 interactive

Available Commands

• ping: Test server connectivity
• list: List available AI models
• interactive: Start interactive chat mode

Client Options

• --host TEXT: DNS server host (default: 127.0.0.1)
• --port INTEGER: DNS server port (default: 5333)
• --api-key TEXT: 10-character API key for authentication (required for interactive mode)

🔑 API Key Generation

Generate a secure 10-character API key:

# OpenSSL
openssl rand -base64 32 | tr -d "=+/" | cut -c1-10

💡 This should be set in the server's API_KEY and set by the client when requestion a chat generation.

🌐 Deployment

To deploy GPT-53 on a subdomain (e.g., gpt53.example.com), follow these steps:

1. DNS Configuration

Set up DNS records for your subdomain:

# A record pointing to your server IP
gpt53.example.com    A    YOUR_SERVER_IP

# NS record to delegate DNS queries to your server (optional, for direct DNS queries)
gpt53.example.com    NS   gpt53.example.com

2. Server Configuration

Configure the server to bind to the appropriate interface:

export HOST=0.0.0.0          # Listen on all interfaces
export PORT=53               # Standard DNS port (requires root/sudo)
export API_KEY=your-secure-10-char-key
export OPENAI_API_KEY=your-openai-api-key

# Start the server
cd server
npm start

3. Firewall Configuration

Ensure your firewall allows DNS traffic:

sudo ufw allow 53/udp
sudo ufw allow 53/tcp

4. Client Usage with Subdomain

Once deployed, clients can connect using your subdomain:

# Python client
gpt53 --host gpt53.example.com --port 53 --api-key your-api-key interactive

# Direct DNS queries
dig @gpt53.example.com TXT "PING"
dig @gpt53.example.com TXT "your-api-key0Hello world"

📋 TODO

• Long Response Support: Bypass response length limits with chunked delivery (return CONTINUE:[chunk_id] when response exceeds limit, client queries GET_CHUNK:[chunk_id]:[part_number])
• Long Message Support: Add START_LONG_MESSAGE and END_LONG_MESSAGE commands to bypass TXT record char limits (use API key as message identifier, server-side message store needed)
• Chat History & Threads: Add chat history persistence and multiple thread support (commands like LIST_THREADS, SELECT_THREAD:[thread_id], NEW_THREAD, server-side store required)
• Function Calling: Add support for function calling in AI responses (return FUNC_CALL:[function_name]:[base64_encoded_args], client executes function and responds with FUNC_RESULT:[base64_encoded_result])