Zro is a private inference API from MoonMath that serves open-weight coding models — starting with MiniMax M3 and GLM-5.2 — from EU-based infrastructure (Finland and France), with zero request retention and no training on customer prompts or completions. It exposes both OpenAI-compatible and Anthropic-compatible endpoints, so existing coding agents can point at it with minimal changes.
The product ships a small @moonmath-ai/zro CLI that logs in once and then launches supported coding harnesses — Claude Code, Codex CLI, Cursor, Cline, OpenCode, Hermes, OpenClaw, and Pi — with temporary session config pointed at Zro’s endpoint. MoonMath positions it as the answer to a tradeoff developers keep hitting: use a closed-source API and give up control over where prompts go, or self-host open models and take on the infrastructure burden yourself.