When one translation isn't enough to say what you mean

I was writing a message to my girlfriend in her first language — something small, just that I'd been thinking about her. Google Translate gave me one option. It was technically correct and read like a form letter.

The problem with literal translation tools isn't accuracy, it's flatness. You get the words but not the weight. There's no difference between what you'd text someone you love and what you'd write in a work email, no note on whether a phrase sounds natural or slightly stiff, no audio so you can hear whether you're about to say something embarrassing.

I built konid to fix that for myself. For any phrase you want to say, it returns three versions — casual, neutral, formal — in the right order, each with a note explaining the register and how the options differ culturally, not just grammatically. It plays the pronunciation through your speakers directly using node-edge-tts, no external API key.

The specific case I kept running into: saying something tender that actually lands. "I missed you" in Spanish has a few shapes — te extrañé, te he echado de menos, others — and which one you use signals something. konid surfaces that signal instead of collapsing it.

It covers 13+ languages including Mandarin, Japanese, Korean, Spanish, French, Arabic, Hindi, and Russian. Runs as an MCP server, so it works inside Claude Code, Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork:

claude mcp add konid-ai -- npx -y konid-ai

Also installs as a ChatGPT app via Developer mode at https://konid.fly.dev/mcp.

MIT licensed. Source: https://github.com/robertnowell/konid-language-learning