Happy slice of mcp pizzaMCP.pizza
archgw

archgw

MCP.Pizza Chef: katanemo

Plano, formerly Arch Gateway, comes from the people who built Envoy and sits in front of the agents and models an application uses. Describe your agents in a settings file and it sends each request to the one that handles it, switches between model providers by name, applies safety and moderation checks before anything reaches a model, and records what happened for later review. Any language or framework works, since it runs alongside your app rather than inside it. Expect Docker, a settings file and your own provider keys.

Coding

Use This MCP server To

Send each request to the agent that handles it Swap AI models without rewriting my app Block unsafe requests before a model sees them See what every request cost and how long it took Add a new agent without touching app code

README

Arch Logo

The AI-native proxy server for agentic applications.

Arch handles the pesky, low-level details like routing user prompts to the right agents or specialized Model Context Protocol (MCP) tools, providing unified access and observability to large language models (LLMs), and quickly clarifying vague user inputs. With Arch, you build faster by focusing on the high-level logic of agents.

Quickstart • Demos • Build agentic apps with Arch • Use Arch as an LLM router • Documentation • Contact

pre-commit rust tests (prompt and llm gateway) e2e tests Build and Deploy Documentation

Overview

Arch - Build fast, hyper-personalized agents with intelligent infra | Product Hunt

Past the thrill of an AI demo, have you found yourself hitting these walls? You know, the all too familiar ones:

  • You go from one BIG prompt to specialized prompts, but get stuck building routing and handoff code?
  • You want use new LLMs, but struggle to quickly and safely add LLMs without writing integration code?
  • You're bogged down with prompt engineering just to clarify user intent and validate inputs effectively?
  • You're wasting cycles choosing and integrating code for observability instead of it happening transparently?

And you think to yourself, can't I move faster by focusing on higher-level objectives in a language/framework agnostic way? Well, you can! Arch Gateway was built by the contributors of Envoy Proxy with the belief that:

Prompts are nuanced and opaque user requests, which require the same capabilities as traditional HTTP requests including secure handling, intelligent routing, robust observability, and integration with backend (API) systems to improve speed and accuracy for common agentic scenarios – all outside core application logic.*

Core Features:

  • 🚦 Routing. Engineered with purpose-built LLMs for fast (<100ms) agent routing and hand-off scenarios
  • ⚡ Tools Use: For common agentic scenarios let Arch instantly clarify and convert prompts to tools/API calls
  • ⛨ Guardrails: Centrally configure and prevent harmful outcomes and ensure safe user interactions
  • 🔗 Access to LLMs: Centralize access and traffic to LLMs with smart retries for continuous availability
  • 🕵 Observability: W3C compatible request tracing and LLM metrics that instantly plugin with popular tools
  • 🧱 Built on Envoy: Arch runs alongside app servers as a containerized process, and builds on top of Envoy's proven HTTP management and scalability features to handle ingress and egress traffic related to prompts and LLMs.

High-Level Sequence Diagram: alt text

Jump to our docs to learn how you can use Arch to improve the speed, security and personalization of your GenAI apps.

archgw FAQ

Do I have to write code to use it?
You describe your agents and models in a small settings file, and the routing, retries and reporting come from the gateway itself.
Which models can it use?
Major providers such as OpenAI and Anthropic, chosen by name or by preference, so you can switch without rewriting your app.
Do I need a key?
Yes, your own provider keys. The routing models it ships with are hosted free of charge for now.
Can I use this to block unsafe or off-topic requests?
Yes — moderation and safety checks run in front of your agents and can reject a request before a model sees it.
How hard is setup?
Developer level: Docker, a settings file and its command line tool.
Is this the same project as Arch Gateway?
Yes — Arch Gateway was renamed Plano, and the code moved to the new repository with it.