I was helping a friend draft a work email in French — a language he's been studying for two years. He typed his sentence into a translation tool, got one answer, and had no way to know if it read as formal, casual, brusque, or warm. He sent it. His French colleague later told him it was technically correct but oddly stiff.
That's the gap literal translation tools leave. One answer, no register, no sense of what it signals to the person reading it.
konid returns three options for anything you want to say, ordered casual to formal, with the register explained and a comparison of the nuance between them. It also plays the pronunciation out loud through your speakers using node-edge-tts — no external API key needed.
The register distinction matters more than people expect. In Japanese, whether you say 「会いたかった」or 「お会いしたかった」or something in between isn't just politeness level — it signals the relationship, the setting, the emotional weight you're putting on the words. A phrasebook gives you the polite form and calls it done. That's fine for ordering coffee; it's not fine for texting your partner or writing to a colleague you want to actually impress.
Same thing in Arabic: the same sentiment in Modern Standard Arabic versus a colloquial dialect reads completely differently to a native speaker. konid flags that.
The tool 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. It also installs as a ChatGPT app via Developer mode using the endpoint https://konid.fly.dev/mcp. Supports 13+ languages including Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, and Hindi.
The intent was never just translation — it's that each query teaches you something about how the language actually works, not just what the words technically mean.
MIT licensed. github.com/robertnowell/konid-language-learning
Top comments (0)