---
title: "LLM gateway overview"
url: "/docs/llm-gateway/overview/"
kind: "concept"
summary: "What the LLM gateway does: one endpoint for several model providers, with failover, token budgets and guardrails. Documentation under construction."
---

> Arion site index: [/llms.txt](/llms.txt)

# LLM gateway overview

What the LLM gateway does: one endpoint for several model providers, with failover, token budgets and guardrails. Documentation under construction.

The LLM gateway runs Arion as one endpoint in front of several model providers, so applications talk to models through the same governed front door as everything else.

> **Under construction.** The LLM gateway ships shortly, and this section is being written alongside it. Check back soon, or watch the [changelog](@/changelog.md).

The pages planned for this section:

- **Providers and formats** — the supported model providers and how request and response formats are translated.
- **Model routing and failover** — routing between models and providers, and failing over when one degrades.
- **Token limits and budgets** — enforcing per-caller token budgets at the gateway.
- **Guardrails** — applying policy to prompts and completions on the data path.
- **Observability** — the metrics and logs the LLM gateway emits.

Routing to self-hosted model servers with an external endpoint picker is a separate feature: the [AI inference gateway](@/docs/ai-inference-gateway/overview.md).