Millwright

Millwright is an open-source router that sits between AI applications and model providers, accepting OpenAI Chat Completions and Anthropic Messages protocols. It routes each request to the lowest-cost model permitted by your policy, supporting OpenAI-compatible APIs, Anthropic, and Amazon Bedrock. The tool lets you define cheap, mid, and frontier model tiers while keeping provider credentials on your infrastructure.

Designed for teams running coding agents and GenAI workloads, Millwright tracks spend per team, maintains prompt-cache affinity across sessions, and provides inspectable routing decisions via response headers and trace logs. It runs as a single Rust binary with optional Docker deployment and SQLite or PostgreSQL backends. According to the creator, portions of the codebase were developed using AI coding agents.