A local-first AI app for learning languages.

Open source. Bring your own model — or none at all.

Free · macOS, Windows, Linux

Built the way language learners actually use these tools.

Local-first by default

Your vocabulary, decks, and reader docs live in a SQLite file on your disk. No account required. No data leaves your machine unless you opt in.

Bring your own AI

Plug in OpenAI, Anthropic, Gemini, OpenRouter, or Ollama. Tokori never sees your keys — they're stored locally and used to call the provider directly.

Click-to-define everywhere

Every word in chat, the reader, or your notes is hover-to-define. Saved words feed your flashcards, your reader-difficulty estimate, and your study progress.

Real spaced repetition

FSRS scheduling under the hood — the same algorithm Anki ships. Stages adapt to how well you know each word, not a fixed Leitner box.

Reader + chat in one place

Drop in a PDF, ebook, or webpage. The reader generates k+1 graded retells, hover-to-define on every word, and saved sentence-mining cards.

Yours forever

Source is on GitHub. Export at any time. If we go away, your data and the app keep working — no servers, no auth, no kill switch.

Bring your own AI. Or run it offline.

Tokori is a UI on top of whichever model you trust. Plug in OpenAI, Anthropic, Gemini, or OpenRouter and the cost is what those providers charge — Tokori takes nothing. Or run Ollama on your machine for fully offline learning. No subscription, no account, no telemetry.

  • OpenAI
  • Anthropic
  • Gemini
  • OpenRouter
  • Ollama
  • Local TTS

Open from the start.

MIT-licensed. Built in Rust + TypeScript. The full source — desktop app, sync server, MCP bridge — is on GitHub. Star it, fork it, audit what runs on your machine.