🚀 Unified LLM API Gateway

---

✨ Overview

Unified LLM API Gateway is a scalable, extensible platform that aggregates and normalises calls to multiple LLM backends (OpenAI, Hugging Face, Groq, Anthropic, Gemini, and more).
It provides a unified API with built-in caching, rate limiting, authentication, logging, metrics, and production-ready deployment manifests for Docker and Kubernetes.

---

---

🏗️ Architecture

• API Gateway (Go):
  • Accepts client requests
  • Handles authentication, routing and request transformation
  • Aggregates/fans-out to LLM backends
• LLM Adapters (microservices):
  • Wrap each provider’s API (OpenAI, Hugging Face, etc.) with a unified internal interface
• Cache Layer:
  • Redis for result caching (prompt+params as cache key)
• Rate Limiter:
  • Redis-based leaky-bucket or token-bucket (shared across instances)
• Auth & Quotas:
  • API keys / JWT, per-key quotas (Redis or DB)
• Observability:
  • Structured logs (JSON), Prometheus metrics, traces
• Deployment:
  • Docker images, Helm charts, Kubernetes manifests, CI builds

---

📁 Monorepo Layout

llm-api-gateway/
├── README.md
├── LICENSE
├── .github/           # CI/CD workflows
├── infra/             # Docker Compose & Kubernetes manifests
│   ├── k8s/
│   └── docker-compose.yml
├── gateway/           # Go API gateway
│   ├── cmd/server/
│   ├── internal/
│   │   ├── handlers/
│   │   ├── adapters/
│   │   ├── cache/
│   │   ├── ratelimit/
│   │   └── metrics/
│   ├── go.mod
│   └── Dockerfile
├── adapters/          # Per-provider adapters (microservices)
│   ├── openai-adapter/
│   └── hf-adapter/
├── admin/             # NestJS admin dashboard (API keys, usage, logs)
│   ├── src/
│   ├── package.json
│   └── Dockerfile
└── tooling/
    └── tests/         # e2e test helpers

---

⚡ Quickstart

1. Start all services:

   docker-compose up --build

2. Gateway API:
   http://localhost:3020/gateway/query
3. Admin Dashboard:
   http://localhost:3040

---

🔌 Supported Providers

• OpenAI (GPT-3.5, GPT-4, GPT-4o, etc.)
• Hugging Face Inference API
• Groq
• OpenRouter
• Anthropic (Claude)
• Gemini (Google)
• More coming soon!

---

🛡️ Features

• Unified API: One endpoint for all LLMs
• Authentication: API key/JWT middleware
• Caching: Redis-based, prompt+params as key
• Rate Limiting: Per-key, Redis-backed
• Logging: Structured, JSON logs
• Monitoring: Prometheus metrics endpoint
• Adapters: Microservices for each provider
• Kubernetes & Docker: Production-ready manifests

---

🧑‍💻 API Usage

Request

POST /gateway/query
Authorization: <your-gateway-api-key>
Content-Type: application/json

{
  "provider": "openai" | "hf" | "groq" | "openrouter" | "anthropic" | "gemini",
  "prompt": "Your prompt here"
}

Response

{
  "cached": false,
  "response": "LLM output"
}

---

🚦 Development Phases

Phase 1: Core Gateway

• Unified /query endpoint
• OpenAI, Hugging Face, Groq, OpenRouter, Anthropic, Gemini support
• Redis caching
• API key authentication
• Rate limiting
• Logging

Phase 2: Adapters & Extensibility

• Per-provider adapters as microservices
• Unified internal API for adapters
• Docker Compose & K8s manifests

Phase 3: Observability & Admin

• Prometheus metrics
• Admin dashboard (NestJS)
• Usage quotas & billing
• Tracing (OpenTelemetry)

Phase 4: Advanced Features (Planned)

• Multi-provider aggregation/fan-out
• Request/response transforms
• Fine-grained quotas & billing
• User/project management
• Webhooks & streaming
• Model selection & fallback
• More adapters (Cohere, Mistral, etc.)

---

📈 Roadmap

• Add more LLM providers & adapters
• Streaming & webhooks support
• Advanced admin features (usage, billing, analytics)
• Helm charts for K8s
• OpenAPI/Swagger docs

---

🤝 Contributing

Contributions are welcome! Please open issues or PRs for bugs, features, or improvements.

---

📄 License

MIT