Model Hotel
A self-hosted, multi-provider AI gateway that exposes one OpenAI-compatible endpoint in front of every configured LLM provider, auto-discovering available models, forming automatic failover groups around shared model names, and storing no prompt data by design.
Model Hotel sits in front of however many LLM providers a user configures and presents them as one OpenAI-compatible endpoint. Models are auto-discovered the moment a provider is added, and again on a schedule if configured; failover groups form automatically around models that share a name across providers, retrying transparently on another provider if one goes down, and no prompt content is ever stored.
It runs as a Docker Compose stack with a web dashboard, printing a one-time admin token to the startup log on first run. Several instances can run behind one client-facing endpoint for high availability, coordinated by a control-plane application that replicates configuration to every member and is load-balanced by Traefik with health checks and automatic failover; members share one master key so encrypted provider keys work across the fleet while each keeps its own admin token. A native Android companion app is also provided.
It is written in Go, self-hosted, and licensed under MIT; the project discloses that AI assistance was used in its development alongside human review at each stage.