Zro

What is Zro?

Zro by MoonMath is a private inference endpoint for coding agents, serving open-weight models like MiniMax M3 and GLM-5.2 from EU infrastructure with zero data retention and no training on customer prompts. It provides an OpenAI-compatible API, a CLI launch tool (@moonmath-ai/zro npm package), and integrations for Claude Code, Codex CLI, Cursor, Cline, OpenCode, Hermes, and OpenClaw, optimized for long-context coding sessions.

Key features

• OpenAI-compatible and Anthropic-compatible Messages API endpoints
• Zero data retention by default — prompts and completions are never stored
• EU-region hosting (Finland, France) for GDPR-friendly inference
• npm launch CLI for one-command setup with supported coding agents
• HyperQuant compression and custom attention kernels for low-latency inference
• Supports AMD, NVIDIA, and TPU hardware
• Models on deck: MiniMax M3, GLM-5.2; DeepSeek V4 Pro and Kimi K2.6 coming
• Plans from $20/month for $60 inference spend, usage packs available

Who it's for

• Teams that need private coding agent inference without sharing prompts with US cloud providers
• Developers who want to swap between open coding models without changing their agent setup
• Builds that require long-context coding sessions (multi-file refactoring, codebase-wide edits)
• Agent frameworks that need an OpenAI-compatible endpoint but can't use OpenAI API

When not to use it

If you need closed-source model quality (GPT-4/Claude class), Zro currently offers only open-weight models. Also not ideal for general-purpose LLM APIs — it's purpose-built for coding agent workloads specifically.

FAQ