# Tiller Router

> One endpoint. One key per client. Steer the models behind it.

- Page: https://www.whatwemade.app/p/tiller-router
- What it is: Web app, for engineering
- Made by: [Ben](https://www.whatwemade.app/u/dellarb)
- Run it yourself: get the code and run your own copy
- Code: https://github.com/dellarb/tiller-router (AGPL-3.0, 1 stars)
- Wants contributors
- Tags: ai-agents, ai-gateway, anthropic-compatible, api-proxy, automatic-failover
- Upvotes: 2, comments: 0

## About

One endpoint. One key per client. Steer the models behind it.

A lightweight, self-hosted LLM router with a control panel built for people who actually change models.

1. **Every tool had its own (bad) model selector.** Some bury it in settings, some want model IDs in config files, some barely support changing models at all. Tiller moves model selection out of the tool and into one control panel — change it without touching the clients. 2. **Cheap and free API access is useful — until it rate-limits.** Tiller gives virtual models an **ordered fallback chain**, so a limited provider is useful without becoming a single point of failure. 3. **Provider API keys were scattered everywhere.** Enter provider credentials once in Tiller; clients only ever receive a Tiller key.

## How it was made

- Claude Code, 20% of the work
- Codex, 90% of the work

Shared on WhatWeMade: https://www.whatwemade.app/p/tiller-router