Project screenshots
Overview
SBproxy is an AI gateway designed to manage API, MCP, agent, and AI model traffic. Built on Cloudflare's Pingora framework, it acts as a reverse proxy and load balancer that sits between client applications and LLM providers or self-hosted models. It unifies access to OpenAI, Anthropic, Gemini, Bedrock, and 60+ other providers via an OpenAI-compatible API, while also allowing users to serve models locally using vLLM or llama.cpp on their own GPUs. The gateway includes enterprise-grade features such as API key management, budgets, guardrails, semantic caching, and a Web Application Firewall (WAF) based on the OWASP Core Rule Set.
Key features
- OpenAI-compatible API for 60+ LLM providers
- Local model hosting via vLLM and llama.cpp
- MCP gateway for tool exposure and federation
- API key, budget, and guardrail management
- Semantic cache
- Web Application Firewall (WAF) using OWASP Core Rule Set
- Rate limiting and load balancing
- Built on Pingora for high-performance proxying
- WebAssembly (Wasm) scripting support via wasmtime
- Kubernetes operator
Requirements, installation and quick start
Via Homebrew (macOS): `brew tap soapbucket/tap && brew install sbproxy`
Via Docker: `docker pull soapbucket/sbproxy:latest`
From Source: `make build-release`
Usage
Model compatibility and use cases
Compatible with OpenAI, Anthropic, Gemini, Bedrock, and 60+ other providers via an OpenAI-compatible API. Supports serving local models using vLLM and llama.cpp inference engines.
License and risk notes
Apache License 2.0
Editorial verification 2026-08-02: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.
Release and maintenance
Not stated in the repository metadata