GitLab Duo Proxy

Local proxy server that exposes GitLab Duo AI models via standard Anthropic and OpenAI APIs — with built-in usage tracking and a real-time management dashboard.

Features

• Dual API Support — Anthropic Messages API (/v1/messages) and OpenAI Chat Completions (/v1/chat/completions)
• 11 Models — Claude (Opus 4.6, Sonnet 4.6/4.5, Haiku 4.5) + GPT-5 variants via GitLab Duo
• Streaming & Non-Streaming — SSE streaming in native format for each API, plus "stream": false for JSON responses
• Proxy API Key — Optional authentication layer to secure the proxy with a single key
• Management Dashboard — Real-time web dashboard at /dashboard with usage stats, per-model metrics, and request history
• Usage Tracking — In-memory request statistics: tokens, latency, success/error rates per model
• Model Aliases — Use sonnet, opus, haiku, codex shortcuts instead of full model names
• Large Payloads — 50MB request body limit for workspace context
• CLIProxyAPI Compatible — Works as an openai-compatibility provider in CLIProxyAPI

Quick Start

# 1. Clone & install
git clone https://github.com/nano-step/gitlab-duo-proxy.git
cd gitlab-duo-proxy
npm install

# 2. Configure
cp .env.example .env
# Edit .env — set your GITLAB_TOKEN and optionally PROXY_API_KEY

# 3. Run
npm run dev

The server starts on http://localhost:3456. Open http://localhost:3456/dashboard to see the management panel.

Configuration

Create a .env file from the example:

cp .env.example .env

Variable	Default	Description
GITLAB_TOKEN	—	Required. GitLab Personal Access Token (scope: api)
GITLAB_INSTANCE_URL	https://gitlab.com	GitLab instance URL
PORT	3456	Server port
HOST	0.0.0.0	Bind address
PROXY_API_KEY	—	Optional. When set, all API requests must include this key

Getting a GitLab Token

1. Go to GitLab → Settings → Access Tokens
2. Create a token with the api scope
3. Add it to your .env file as GITLAB_TOKEN

API Endpoints

Health Check

curl http://localhost:3456/health
# {"status":"ok"}

List Models

curl http://localhost:3456/v1/models

Anthropic Messages API

curl -X POST http://localhost:3456/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: $PROXY_API_KEY" \
  -d '{
    "model": "sonnet",
    "messages": [{"role": "user", "content": "Hello"}],
    "max_tokens": 200,
    "stream": false
  }'

OpenAI Chat Completions API

curl -X POST http://localhost:3456/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $PROXY_API_KEY" \
  -d '{
    "model": "sonnet",
    "messages": [
      {"role": "system", "content": "You are helpful."},
      {"role": "user", "content": "Hello"}
    ],
    "max_tokens": 200,
    "stream": false
  }'

Set "stream": true for Server-Sent Events streaming.

Proxy Authentication

When PROXY_API_KEY is set in .env, all requests to /v1/messages and /v1/chat/completions must include the key via one of:

Method	Header
Anthropic style	x-api-key: <your-proxy-key>
OpenAI style	Authorization: Bearer <your-proxy-key>

The /health, /v1/models, and /dashboard endpoints remain public.

If PROXY_API_KEY is not set, the proxy operates without authentication (backward compatible).

Management Dashboard

Access the real-time dashboard at:

http://localhost:3456/dashboard

The dashboard displays:

• Summary Cards — Total requests, success rate, total tokens, average latency
• Model Table — Per-model breakdown with request count, token usage, average latency, and error count (color-coded by provider: Anthropic/OpenAI)
• Daily Activity Chart — Bar chart showing requests per day over the last 14 days
• Recent Requests — Last 20 requests with timestamp, model, endpoint, status, tokens, and latency

Data auto-refreshes every 5 seconds. No external dependencies — everything is self-contained.

Dashboard API

Raw JSON data is available at:

# Usage statistics snapshot
curl http://localhost:3456/dashboard/api/stats

# Model list with per-model stats
curl http://localhost:3456/dashboard/api/models

Models

Duo Model	Provider	Backend	Aliases
duo-chat-opus-4-6	Anthropic	claude-opus-4-6	opus, opus-4-6
duo-chat-sonnet-4-6	Anthropic	claude-sonnet-4-6	sonnet-4-6
duo-chat-opus-4-5	Anthropic	claude-opus-4-5-20251101	opus-4-5
duo-chat-sonnet-4-5	Anthropic	claude-sonnet-4-5-20250929	sonnet, sonnet-4-5
duo-chat-haiku-4-5	Anthropic	claude-haiku-4-5-20251001	haiku, haiku-4-5
duo-chat-gpt-5-1	OpenAI	gpt-5.1-2025-11-13	gpt-5.1, gpt-5-1
duo-chat-gpt-5-2	OpenAI	gpt-5.2-2025-12-11	gpt-5.2, gpt-5-2
duo-chat-gpt-5-mini	OpenAI	gpt-5-mini-2025-08-07	gpt-5-mini
duo-chat-gpt-5-codex	OpenAI	gpt-5-codex	gpt-5-codex
duo-chat-gpt-5-2-codex	OpenAI	gpt-5.2-codex	codex, gpt-5.2-codex
duo-chat-gpt-5-3-codex	OpenAI	gpt-5.3-codex	gpt-5.3-codex

CLIProxyAPI Integration

This proxy works as an openai-compatibility provider in CLIProxyAPI. Add to your config.yaml:

openai-compatibility:
  - name: "gitlab-duo"
    base-url: "http://localhost:3456/v1"
    api-key-entries:
      - api-key: "your-proxy-api-key"
    models:
      - name: "duo-chat-sonnet-4-5"
        alias: "gitlab-sonnet"
      - name: "duo-chat-opus-4-6"
        alias: "gitlab-opus"
      - name: "duo-chat-haiku-4-5"
        alias: "gitlab-haiku"
      - name: "duo-chat-gpt-5-2-codex"
        alias: "gitlab-codex"

Running as a Service

Using start.sh (daemon mode)

chmod +x start.sh
./start.sh          # Start in background
./start.sh status   # Check status
./start.sh logs     # View logs
./start.sh stop     # Stop

Using npm

npm run dev          # Development (tsx, auto-reload)
npm run build        # Compile TypeScript
npm start            # Production (compiled JS)

Architecture

┌─────────────┐     ┌─────────────────────┐     ┌─────────────┐
│   Client    │────▶│   GitLab Duo Proxy  │────▶│  GitLab AI  │
│ (CCS/curl)  │◀────│    localhost:3456    │◀────│   Gateway   │
└─────────────┘     └─────────────────────┘     └─────────────┘
                              │
               ┌──────────────┼──────────────┐
               ▼              ▼              ▼
         /v1/messages   /v1/chat/      /dashboard
         (Anthropic)    completions    (Management UI)
                        (OpenAI)

License

MIT