Deploy and Host Bifrost on Railway

Bifrost is a high-performance AI gateway. It exposes one OpenAI-compatible API in front of 20+ providers (OpenAI, Anthropic, AWS Bedrock, Google Vertex, Groq, Ollama and more) and adds automatic failover, load balancing, semantic caching, virtual keys and budgets, with a web UI for configuration and analytics.

About Hosting Bifrost

This template runs the official maximhq/bifrost:latest image on port 8080 with a persistent volume at /app/data for its config and log databases. The template generates a random BIFROST_SETUP_TOKEN and BIFROST_ENCRYPTION_KEY (both in the service Variables tab). Until the first admin account exists, every API call must carry the setup token, so a stranger cannot claim your gateway: open the URL, create your admin account (enter the setup token if the UI asks for it), then add your provider API keys in the UI. Provider keys are stored encrypted with the generated encryption key; do not change that variable afterwards.

Inference endpoints require authentication by default once the admin account exists; use a virtual key or the admin credentials for your clients.

Common Use Cases

One endpoint for several LLM providers with fallback between them
Per-team virtual keys with budgets and rate limits
Caching and observability in front of an existing OpenAI-style app

Dependencies for Bifrost Hosting

One service with a volume; the LLM provider API keys you add in the UI

Deployment Dependencies

Why Deploy Bifrost on Railway?

Railway gives you a public HTTPS URL, a persistent volume, generated secrets and automatic restarts with no server to maintain.