Skip to content
GPROXY

What is GPROXY?

GPROXY is a high-performance LLM proxy server written in Rust. It exposes a unified, OpenAI / Anthropic / Gemini compatible HTTP surface on top of many upstream providers, while adding the primitives you need to run it as a shared service: users, API keys, model permissions, rate limits, cost quotas, usage logging, and an embedded browser console.

It ships as a single static binary (with an embedded React console) and an optional Rust SDK for developers who want to reuse the engine in their own applications.

What it is good at

Section titled “What it is good at”
  • Fanning out to many upstreams from one endpoint. A single GPROXY instance can route to OpenAI, Anthropic, Vertex / Gemini, DeepSeek, Groq, OpenRouter, NVIDIA, Claude Code, Codex, Antigravity, custom OpenAI-compatible endpoints, and more — each configured as an independent provider.
  • Multi-tenant access control. Issue API keys to individual users, gate them with glob-style model permissions, apply RPM / RPD / token rate limits per model pattern, and enforce USD-denominated quotas with a reconciler running in the background.
  • Cross-protocol translation. A client speaking the OpenAI Chat Completions format can be routed to an Anthropic or Gemini upstream (and vice versa) through the protocol transform layer.
  • Same-protocol passthrough. When the client and upstream speak the same protocol, GPROXY forwards bytes with minimal parsing for low-overhead, high-throughput operation.
  • Operational visibility. Structured upstream / downstream logs (with optional body capture), per-request usage accounting, model health tracking, and a web console that surfaces all of it.

What it is not

Section titled “What it is not”
  • It is not a model host. GPROXY does not run inference itself; it talks to real upstream providers over HTTP.
  • It is not a load balancer for web traffic generally. It understands LLM protocols (OpenAI, Claude, Gemini) and is optimized for them.
  • It does not ship a managed UI behind SSO. Authentication for the embedded console is a username + password that issues a bearer session token; integrate it behind your own reverse proxy if you need more.

Core concepts at a glance

Section titled “Core concepts at a glance”
ConceptWhat it means in GPROXY
ProviderA configured upstream (name + channel + settings + credentials).
ChannelThe code that speaks a specific upstream protocol (OpenAI, Anthropic, Gemini, …).
ModelA forwardable model id on a provider. May carry pricing.
AliasA friendly name that resolves to a real (provider, model) pair.
UserAn account with one or more API keys, permissions, limits, and quota.
PermissionA (user, provider, model_pattern) tuple granting routable access.
Rate limitRPM / RPD / token ceilings scoped to a user + model pattern.
QuotaA cost ceiling (USD) enforced across all usage for a user.

Where to go next

Section titled “Where to go next”