GitHub Repository

High-performance self-hosted AI gateway (OpenAI-compatible) with routing, retries, and streaming

19 starsGo

LunarGate – a self-hosted OpenAI-compatible LLM gateway

Name: LunarGate – a self-hosted OpenAI-compatible LLM gateway
Availability: InStock
Author: jmartenka

by jmartenka·Apr 10, 2026·3 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidShip ItNiche Gem

Go gateway with circuit breakers, but auth isn't production-ready yet.

Strengths

•10-12 MB binary footprint is genuinely small for a gateway with streaming support
•Complexity-aware routing and weighted provider rules go beyond basic load balancing
•Hot-reloadable YAML config without restarting the gateway process

Weaknesses

•Author admits inbound auth and security is not production-complete
•LiteLLM and similar gateways already serve this exact use case

Post Description

Hi HN — I built LunarGate, a self-hosted OpenAI-compatible LLM gateway written in Go.

It exists because once you add multiple model providers, retries, fallbacks, routing, and observability logic start leaking into app code.

LunarGate moves that into one gateway layer.

Current features: - provider/model routing - retries and fallbacks - circuit breakers - optional caching and rate limiting - hot-reloadable YAML config - optional observability - complexity-aware routing

Important caveat: inbound auth/security is not production-complete yet, so it should run behind an auth-enforcing proxy/gateway. I’m actively working on that now.

I also have a long roadmap, but I’m trying to figure out what matters most in practice. If you’ve run something like this before, I’d especially love to know: what 1–2 things would you need before trusting a gateway like this in production?

It’s open source: https://github.com/lunargate-ai/gateway

Docs / site: https://docs.lunargate.ai/ https://lunargate.ai