Arch Introduction
Arch is an intelligent Layer 7 gateway built on Envoy Proxy that provides secure handling, robust observability, and seamless integration of prompts with APIs for building fast, robust, and personalized AI agents.
View MoreWhat is Arch
Arch is an open-source intelligent prompt gateway engineered with purpose-built LLMs to handle critical tasks related to prompt processing and API integration. Built by the core contributors of Envoy proxy, it treats prompts as nuanced user requests that require the same capabilities as traditional HTTP requests. The platform is designed to protect, observe, and personalize LLM applications like agents, assistants, and co-pilots while integrating them with backend APIs.
How does Arch work?
Arch operates by handling undifferentiated tasks in prompt processing through several key mechanisms. It uses specialized LLMs for detecting and rejecting jailbreak attempts, intelligently calling backend APIs to fulfill user requests, and managing prompt routing between upstream LLMs. The system runs alongside application servers and leverages Envoy's HTTP management capabilities for handling ingress and egress traffic. It includes purpose-built LLMs for function/API calling and parameter extraction, while centralizing prompt guardrails through its Prompt Guard feature. The platform implements the W3C Trace Context standard for comprehensive request tracing and provides robust traffic management with features like smart retries and automatic cutover.
Benefits of Arch
Using Arch brings multiple advantages to AI application development. It eliminates the need to write custom code for prompt security by providing centralized prompt guardrails. The platform offers improved performance through fast, cost-effective, and accurate prompt-based tasks, while ensuring continuous availability through resilient upstream connections and smart traffic management. Organizations benefit from comprehensive observability with standardized metrics for monitoring latency, token usage, and error rates. Additionally, its integration with Envoy Proxy provides proven scalability and reliability, making it easier for developers to build and maintain secure, high-performing AI applications.
Popular Articles
12 Days of OpenAI Content Update 2024
Dec 18, 2024
Midjourney Launches Pinterest-like Moodboards Following Patchwork Release
Dec 17, 2024
Midjourney Unveils Patchwork: A New Era in Collaborative Worldbuilding
Dec 17, 2024
Google Launches Whisk: Revolutionary AI Image Generator Remixes Three Images into One
Dec 17, 2024
View More