Zro

Zro is a private inference endpoint built specifically for coding agents. It lets you run open-weight models like MiniMax M3 and GLM-5.2 on EU-hosted infrastructure, with a promise of zero data retention and no training on your prompts or completions. If you're building or using AI coding tools that need to keep code and queries private, Zro gives you a fast, compatible API that works with the agents you already use.

Under the hood, Zro uses MoonMath's HyperQuant compression and custom attention kernels to handle long-context, multi-turn coding sessions efficiently. It supports both OpenAI-compatible and Anthropic-compatible API shapes, so you can point existing clients at Zro's base URL without rewriting code. The @moonmath-ai/zro npm package makes it easy to launch supported tools like Claude Code, Codex CLI, Cursor, and Cline with a single command. Inference runs in Finland and France, and billing starts at $20/month for $60 of spend.

This tool is for developers who need privacy-compliant inference for coding agents — whether you're building an internal assistant, running automated code reviews, or experimenting with open models. It's also useful for teams in regulated industries who can't send code to US-based cloud providers. Zro keeps your data in the EU and never uses it for training, so you get the speed of a modern inference endpoint without the privacy trade-offs.