AI Relay

Bypass network restrictions and access ChatGPT from isolated networks. A relay system that queues prompts from an internal server (with limited internet access) to a local bridge machine (with open internet), then delivers AI responses back.

---

What is AI Relay?

AI Relay is a queue-based relay system that enables ChatGPT access in environments where:

• The internal network has restricted or no access to international internet
• A separate machine (e.g., laptop with Starlink or VPN) has open internet access

Instead of the internal server calling OpenAI directly (which would fail), a worker runs on the machine with internet. It polls the internal server for pending prompts, sends them to the OpenAI API, and returns the responses. This effectively bypasses filtering by routing AI traffic through the bridge machine.

Use Cases

• Corporate networks with strict outbound filtering
• Regions where AI services are blocked or throttled
• Air-gapped or semi-isolated networks with one bridge to the internet
• Educational or research environments with limited external access

---

Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                     RESTRICTED / INTERNAL NETWORK                        │
│                                                                         │
│   Users submit prompts via web UI                                       │
│   ┌──────────────┐         ┌─────────────────────┐                     │
│   │   Browser    │ ──────► │   Internal Server    │                     │
│   │   (Users)    │ ◄────── │   (Go + SQLite)     │                     │
│   └──────────────┘         │   • No OpenAI call  │                     │
│                            │   • Queue in DB     │                     │
│                            └──────────┬──────────┘                     │
│                                       │                                 │
│                                       │ Polling (outbound only)         │
└───────────────────────────────────────┼─────────────────────────────────┘
                                        │
                                        ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                     BRIDGE MACHINE (Open Internet)                        │
│                                                                         │
│   ┌─────────────────────┐                    ┌─────────────────────┐  │
│   │   Worker (Go)        │ ──── HTTPS ──────► │   OpenAI API         │  │
│   │   • Fetches jobs     │ ◄─── Response ──── │   (ChatGPT)          │  │
│   │   • Sends to OpenAI  │                    └─────────────────────┘  │
│   │   • Returns results │                                               │
│   └─────────────────────┘                                               │
└─────────────────────────────────────────────────────────────────────────┘

Key design: The internal server never initiates connections to the outside. Only the worker (on the bridge machine) polls the server and talks to OpenAI. This works even when inbound connections are blocked.

---

Features

Feature	Description
Queue-based	Prompts are stored in SQLite; worker processes them in order
Polling model	Worker fetches jobs periodically—no inbound ports needed on internal server
Session auth	Secure login with bcrypt and session cookies
RTL / Persian UI	Right-to-left support, Vazirmatn font
Rate limiting	Per-user limits to prevent abuse
CSRF protection	Forms protected against cross-site request forgery
Atomic job claiming	Prevents duplicate processing by multiple workers
Docker support	One-command deployment with Docker Compose

---

Quick Start

Prerequisites

• Go 1.21+
• OpenAI API key
• (Optional) Docker & Docker Compose

1. Clone

git clone https://github.com/YOUR_USERNAME/ai-relay.git
cd ai-relay

2. Configure

cp .env.example .env
# Edit .env: set WORKER_TOKEN, DEFAULT_PASSWORD, OPENAI_API_KEY

3. Run Internal Server (on restricted network)

go run ./cmd/server
# Or with Docker:
# docker-compose up -d

Server runs at http://localhost:8080 (or :8081 if port 8080 is in use).

4. Run Worker (on bridge machine with internet)

# .env on bridge machine:
# INTERNAL_SERVER_URL=http://internal-server-ip:8080
# WORKER_TOKEN=<same as server>
# OPENAI_API_KEY=sk-...

go run ./cmd/worker

5. Use

1. Open http://<server-ip>:8080/login
2. Log in (default: admin / admin123)
3. Submit prompts; worker fetches and processes them; responses appear in the dashboard

---

Configuration

Server (.env)

Variable	Description	Default
SERVER_PORT	HTTP port	8080
SESSION_SECRET	Session encryption key	Required
WORKER_TOKEN	Shared secret for worker API	Required
DEFAULT_USERNAME	Initial user	admin
DEFAULT_PASSWORD	Initial password	changeme123
DATABASE_PATH	SQLite path	./data/ai_relay.db

Worker (.env)

Variable	Description	Default
INTERNAL_SERVER_URL	Internal server URL	Required
WORKER_TOKEN	Must match server	Required
OPENAI_API_KEY	OpenAI API key	Required
OPENAI_MODEL	Model name	gpt-4o-mini
WORKER_POLL_INTERVAL_SECONDS	Poll interval	5

---

Project Structure

ai-relay/
├── cmd/
│   ├── server/     # Internal server (web + API)
│   └── worker/     # Bridge worker (polls + OpenAI)
├── internal/
│   ├── config/     # Config loading
│   ├── db/         # SQLite layer
│   ├── handlers/   # HTTP handlers
│   ├── middleware/ # Auth, CSRF, rate limit
│   ├── models/     # Data models
│   ├── openai/     # OpenAI client (worker only)
│   └── services/   # Worker HTTP client
├── web/
│   ├── static/     # CSS
│   └── templates/  # HTML
├── docker-compose.yml
├── Dockerfile
└── .env.example

---

Docker

# Build and run
docker-compose up -d --build

# Default port: 8081 (8080 if available)
# Data persisted in ./data/

---

Security

• Never commit .env — it contains secrets. Use .env.example as a template.
• If .env was accidentally committed: git rm --cached .env then commit the removal.
• Set SESSION_SECRET, WORKER_TOKEN, DEFAULT_PASSWORD via environment — no defaults in code.
• Use HTTPS in production.
• Generate tokens: openssl rand -base64 32
• Restrict worker API access by IP if possible.

---

License

MIT License. See LICENSE for details.