The proxy built for
AI infrastructure
One proxy between your apps and every AI provider. Route traffic, manage keys, enforce guardrails — all from a single config file. Built in Rust for performance and safety.
Why teams choose Praxis
Route to Any AI Provider
Send requests to OpenAI, Anthropic, or your own models. Praxis reads the model name, injects the right API key, and routes to the right upstream — your app never touches a credential.
Built for Claude Code & Codex
Native MCP gateway — aggregate tools from multiple servers behind one endpoint. Praxis handles session management, credential injection, and tool-call routing so your AI coding agents work securely at scale.
Build Pipelines, Not Middleware
Stack rate limiting, auth, guardrails, and routing in any order using a simple YAML config. 50+ built-in filters ship ready to use, and you can write your own in Rust.
Security Without the Setup
TLS everywhere, zero unsafe code, CORS and rate limiting out of the box. Praxis ships locked down so you don't have to bolt security on after the fact.
Ship Changes Without Downtime
HTTP/1.1, HTTP/2, gRPC, WebSocket, TCP — all supported. Hot-reload your config at runtime without dropping a single connection.
Understand the architecture
Interactive visual guides that walk through how Praxis works with animated diagrams.
Praxis Core
How requests flow through filter pipelines, how chains compose, and how config reloads without dropping a connection.
Read the Core guide →Praxis AI
How the AI gateway classifies requests, routes to providers, injects credentials, and speaks MCP and A2A.
Read the AI guide →Praxis Grid
How the distributed control plane converges state with CRDTs, discovers sites with SWIM, and scores routes by locality.
Read the Grid guide →Deploy in minutes, not weeks.
MIT licensed. One binary, one config file, production-ready.