Skip to content

Interactive Playground

Try Multilingual 1.0 directly in your browser — no installation required.

Multilingual is the only AI programming platform where the same agent logic, reactive UI, and semantic search pipeline is idiomatic in any of 17 human languages.

What you can do

  • Write programs using keywords from any of the 17 supported languages (English, French, Spanish, German, Hindi, Arabic, Bengali, Tamil, Chinese, Japanese, Italian, Portuguese, Polish, Dutch, Swedish, Danish, Finnish)
  • Use Core 1.0 language features: fn, |>, ?, ~=, observe var, @agent, @tool, prompt, think, stream, embed
  • Click ▶ Run (or press Ctrl+Enter) to execute the code
  • Inspect the Semantic IR — the typed intermediate representation that drives all backends
  • Explore the Generated Python and WAT / WASM views

Deployment models

Live playground (Pyodide)

Ideal for exploration, teaching, and instant feedback. First load takes a few seconds while the WASM runtime (~12 MB) initializes. Subsequent visits are fast thanks to browser caching.

→ Open the Interactive Playground

Prebuilt bundle (ahead-of-time)

For production web apps, compile ahead of time with build-wasm-bundle, ship module.wasm, and load it with the generated host_shim.js — no runtime download required.

multilingual build-wasm-bundle my_app.ml --out-dir dist/wasm

Live demos using this approach are in examples/browser.

Core 1.0 example: the same agent in three languages

=== English 

fn summarize(text: str) -> str uses ai:
    let result = think @claude-sonnet:
        Summarize the following in one sentence.
        text: text
    return result.conclusion

@agent(model=@claude-sonnet)
fn research_agent(query: str) -> str uses ai, net:
    let answer = summarize(query)
    return answer

=== Français 

fn résumer(texte: str) -> str utilise ai:
    soit résultat = réfléchir @claude-sonnet:
        Résumer en une phrase.
        texte: texte
    retourner résultat.conclusion

@agent(modèle=@claude-sonnet)
fn agent_recherche(requête: str) -> str utilise ai, réseau:
    soit réponse = résumer(requête)
    retourner réponse

=== 日本語 

関数 要約する(テキスト: str) -> str 使う ai:
    変数 結果 = 考える @claude-sonnet:
        一文で要約する。
        テキスト: テキスト
    返す 結果.結論

@エージェント(モデル=@claude-sonnet)
関数 調査エージェント(クエリ: str) -> str 使う ai, ネット:
    変数 回答 = 要約する(クエリ)
    返す 回答

All three programs have identical semantics — the surface language is purely cosmetic.

Reactive counter example

observe var count: int = 0

fn increment():
    count = count + 1

canvas counter_view {
    observe count
}

on count.change:
    render counter_view with count

Semantic matching example

fn classify_intent(input: str) -> str:
    if input ~= "hello":
        return "greeting"
    if input ~= "thank you":
        return "gratitude"
    return "other"

Classic multilingual example

=== English 

let a = 10
let b = 3
print("Sum:", a + b)
for i in range(1, 5):
    print(i, "squared =", i * i)

=== Français 

soit a = 10
soit b = 3
afficher("Somme:", a + b)
pour i dans intervalle(1, 5):
    afficher(i, "au carré =", i * i)

=== Deutsch 

sei a = 10
sei b = 3
ausgeben("Summe:", a + b)
für i in bereich(1, 5):
    ausgeben(i, "Quadrat =", i * i)

=== 日本語 

変数 a = 10
変数 b = 3
表示("合計:", a + b)
毎 i 中 範囲(1, 5):
    表示(i, "の2乗 =", i * i)