pi-language-tutor

Install

pi install npm:pi-language-tutor

That’s the only required step. Defaults (learning English, native Simplified Chinese) work out of the box. Another language? Type /lang or run /lang native ja.

pi install git:github.com/mackt/pi-language-tutor

Or symlink a clone into pi’s global extensions directory (auto-discovered via pi.extensions, hot-reloads with /reload):

git clone https://github.com/mackt/pi-language-tutor.git
ln -s "$(pwd)/pi-language-tutor" ~/.pi/agent/extensions/pi-language-tutor

Features

✏ Writing check

Prompt in the language you’re learning. While the agent works, a panel explains each mistake in your native language — fix, why, and a more natural whole sentence.

✏ Writing tutor

Prompt in your native language because the thought came faster that way. The tutor teaches the natural whole-sentence form in the learning language, the key words, and the grammar that carries it.

🗂 Flashcard review

Words taught by the Writing tutor are saved automatically. Run /flashcards to study them with an Anki-style flow; an FSRS scheduler decides when each card comes back.

🌐 Bilingual cards

After a reply, press alt+t (macOS: ⌥T) or run /translate. Each paragraph is followed by its translation — immersive-translate style, with short code blocks kept intact.

Try this first

1. Start pi and send a prompt in your learning language:

   when agent anwser me, I want translate it, it have three feature
   

   The agent answers as usual. A ✏ Writing check panel appears above the editor with fixes and native-language explanations.

2. Write a prompt in your native language instead:

   我想重构这个函数但是不知道怎么下手
   

   A ✏ Writing tutor panel appears: a natural whole-sentence rendering, key vocabulary, and grammar.

3. Words from the tutor are now flashcards. Run /flashcards, show the answer, then rate yourself Again / Hard / Good / Easy.

4. When the agent finishes, press alt+t (on macOS, enable Option-as-Meta in your terminal, or run /translate). The reply re-renders as a bilingual card.

5. Prefer auto-translate on every final response?

   /lang auto on
   

That’s enough to start.

Design principles

Nothing ever blocks	Your message goes to the agent immediately; the review runs in parallel. A clean message shows no panel at all.
Two panels, never both	Learning language → Writing check. Native language → Writing tutor. One LLM call decides — they never both fire.
Nothing pollutes the conversation	Translation cards live only in your terminal — never sent back to the LLM, no context cost.
You control the spend	Features use your session model by default; point them at a cheaper one with /lang model.

Settings

Type /lang for the interactive menu, or set options directly:

Command	What it does
/translate or alt+t	Translate the last assistant response
/flashcards	Review flashcards captured from the Writing tutor
/lang	Interactive settings menu
/lang check off | on | context	Writing check & tutor mode (context sees the conversation; /lang on/off still work)
/lang tutor on | off	Keep / drop the writing tutor alone
/lang auto on | off	Auto-translate every final response
/lang native <code>	Native language — translations & explanations (zh-CN, ja, …)
/lang learning <code>	Language you’re practicing (en, fr, …)
/lang model [model]	Model for this extension
/lang model default	Follow the session model
/lang context on | off	Translations with full session context (off by default)

Configuration

Settings persist in ~/.pi/agent/language-learn.json.

{
  "learning": "en",
  "native": "zh-CN",
  "model": "openai/gpt-4o-mini",
  "check": "on",
  "tutor": true,
  "auto": false,
  "context": false
}

model defaults to the session model. tutor defaults to on.

Flashcards are stored in ~/.pi/agent/flashcards.json. Review limits and FSRS desired retention are stored in ~/.pi/agent/flashcards-settings.json.

Advanced

Skip heuristics, check vs tutor rules, flashcard scheduling, bilingual card details, custom providers, and context-mode cost trade-offs:

→ Advanced behavior

Development

npm install
npm run check   # typecheck
npm test        # unit tests
npm run lint
npm run fmt:check

Layout, scripts, and how product screenshots are produced:

→ Development guide

Contributions welcome via PR. Conventions live in AGENTS.md.

License

MIT