Put it inside your app in an afternoon.

Node, Electron or a backend.

Import the factory and own the rendering. Nothing fetches or draws your artwork for you.

import { createTinkySpeak } from 'tinkyspeak-runtime';

const session = createTinkySpeak({
  provider: { kind: 'tinkymind' },
  language: 'en', partnerLanguage: 'en',
  profile: { age: 7, level: 'sentence', choices: 6 },
  symbols: choice => yourArtworkFor(choice),
});
const state = await session.hear({ text: 'Would you like a drink?' });
renderYourTiles(state.choices);

// In YOUR user-selection handler:
const selected = session.select({ choiceId: chosenTileId });
const pending  = await session.speak({ draftId: selected.draft.id });
// Your app delivers pending.speech.text in pending.speech.language.
session.confirmSpeech({ speechId: pending.speech.id, outcome: 'spoken' });

Any language, any platform.

Run tinkyspeak serve and the whole engine is a local JSON API on 127.0.0.1:7437. The included fetch client has no Node imports, so it bundles straight into a web app.

import { createTinkySpeakClient } from 'tinkyspeak-runtime/client';

const api     = createTinkySpeakClient({ url: 'http://127.0.0.1:7437' });
const session = await api.createSession({ language: 'en', partnerLanguage: 'en' });
const state   = await session.hear({ text: 'Are you hungry?' });
renderYourTiles(state.choices);

const selected = await session.select({ choiceId: chosenTileId });
const pending  = await session.speak({ draftId: selected.draft.id });
await yourSpeechEngine(pending.speech.text, pending.speech.language);
await session.confirmSpeech({ speechId: pending.speech.id, outcome: 'spoken' });

Selection, draft and speech IDs prevent stale or repeated actions. Concurrent actions return 409 session_busy. Errors come back as {"error":{"code":"...","message":"..."}}. No conversation request bodies are logged.

The boundary is deliberate.

Bring the model you already use.

The runtime ships adapters, not weights.

{ kind: 'tinkymind' }                       // installed local child model; English only
{ kind: 'gemini',   model: 'YOUR_MODEL', apiKeyEnv: 'GEMINI_API_KEY' }
{ kind: 'ollama',   model: 'YOUR_INSTALLED_MODEL' }
{ kind: 'compatible', endpoint: 'https://your-model.example/v1', model: 'YOUR_MODEL' }
{ kind: 'board',    tiles: yourAuthoredTiles }  // no generative model at all