Plano (formerly Arch Gateway)
An open-source AI gateway built on Envoy. Self-host agent orchestration, route LLM requests, and capture OpenTelemetry traces under Apache 2.0.
Plano, formerly Arch Gateway, is a self-hosted AI gateway for developers building applications with multiple agents or model providers. It puts routing, guardrails and request tracing in a separate service, so each agent doesn't need its own implementation of that infrastructure. It's open source under Apache 2.0.
Built on Envoy and written primarily in Rust, Plano works with agents in any language or framework that expose an OpenAI-compatible chat completions API. Its orchestration uses purpose-built models such as Plano-Orchestrator to choose agents and coordinate their work within a conversation. Developers can add agents without rewriting application routing logic, or use the LLM gateway independently of agent orchestration.
Model requests can target a specific model, a named alias or an automatic choice based on preferences. Plano handles provider differences through a unified API, with OpenAI and Anthropic shown as supported providers. It also captures agent activity signals and OpenTelemetry traces, metrics and logs without requiring tracing code in each agent. Filter chains provide hooks for jailbreak protection, moderation policies and memory.
The proxy and Plano's routing models can run locally. A hosted service also runs Plano and its model family in the US-central region; requests using that service or external model providers go beyond your local deployment. Local hosting of the gateway doesn't make cloud model calls local.
