I was drafting a work email in French last year — intermediate level, good enough to get through tasks, not good enough to know when I sounded stiff. Google Translate gave me one sentence. No indication of whether it was formal, casual, or somewhere in between. I sent it, got a reply that matched my register exactly, and realized I'd accidentally set the tone for the whole relationship as "overly professional stranger."
The problem isn't that translation tools are wrong. It's that they give you one answer when natural language has a spectrum. "I missed you today" in Japanese isn't a single phrase — it's a choice between something you'd say to a close friend, something you'd say to a partner, and something tender enough to feel vulnerable. The tool should show you that choice, not make it for you.
konid returns three options per query, ordered casual to formal, with each register labeled and explained. For Mandarin: 我今天想你了 (warm, everyday), 今天一直在想你 (intimate, carries more weight), 今天甚是想念你 (formal-literary, the kind of phrasing you'd read, not speak). The cultural note explains why the third one reads as archaic in conversation. You hear all three out loud through built-in audio — no API key, uses node-edge-tts directly through your speakers.
The angle I kept returning to while building this: the register you choose signals the relationship you think you're in. If you pick the wrong one, you either sound cold when you mean warm, or presumptuous when you mean polite. A phrasebook can't flag that. A literal translation tool won't.
13 languages supported — Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and more.
It runs as an MCP server, so it works inside Claude Code (claude mcp add konid-ai -- npx -y konid-ai), Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. Also installs as a ChatGPT app via Developer mode using the endpoint https://konid.fly.dev/mcp.
MIT licensed. github.com/robertnowell/konid-language-learning