@original-land/original-connect

Framework-agnostic TypeScript SDK for Original Connect — the OAuth 2.1 / OpenID Connect identity layer for the Original AI ecosystem. Log users in, chat with AI agents, add voice and file attachments — without ever exposing API keys to the user.

Install

npm install @original-land/original-connect

This is a public package on the npm registry — no authentication or .npmrc is required.

Which prompt should I use?

| Scenario | Prompts to use | |---|---| | My app has no backend (browser / frontend only) | getting-started.md + chat-browser-app.md | | My app has a backend (Next.js, NestJS, SvelteKit…) | getting-started.md + chat-backend-app.md | | I want to sell access to an agent (paid / marketplace) | getting-started.md + purchasable-agent.md | | I want to add voice / audio | Also add: voice-enabled-app.md | | I want to add file uploads | Also add: file-attachments.md |

Quick start (browser app with a fixed bot)

import { createBrowserClient } from '@original-land/original-connect/browser';
import { Scopes, agentChatScope, buildScope } from '@original-land/original-connect';

const BOT_ID = '<botId>';

const oa = createBrowserClient({
  clientId: '<your-client-id>',
  issuer: 'https://ai.original.land',
  redirectUri: '<your-redirect-uri>',
  responsesApiUrl: 'https://ai-api.original.land',
});

// 1. Login
oa.login({
  scope: buildScope([Scopes.openid, Scopes.profile, agentChatScope(BOT_ID)]),
});

// 2. Handle the callback (on your redirect page)
if (window.location.search.includes('code=')) {
  await oa.handleCallback();
}

// 3. Chat
const result = await oa.streamResponse(BOT_ID, {
  input: [{ type: 'message', role: 'user', content: 'Hello!' }],
}, {
  onTextDelta: (delta) => console.log(delta),
});

See prompts/ for complete guides.

Prompts

Ready-to-paste prompts for coding agents live in prompts/:

| Prompt | Description | |---|---| | getting-started.md | Registration, scopes, security rules, end-to-end flow | | chat-browser-app.md | Browser-only app (PKCE, no backend) | | chat-backend-app.md | Backend app (confidential client) | | purchasable-agent.md | Paid agent / marketplace | | voice-enabled-app.md | Add-on: STT, TTS, Omni v3, live voice sessions | | file-attachments.md | Add-on: file upload and chat with documents |

Canonical domains

| What | URL | |---|---| | Issuer (OAuth) | https://ai.original.land | | Discovery | https://ai.original.land/.well-known/openid-configuration | | Registration | https://ai.original.land/oauth/register | | Responses API (chat) | https://ai-api.original.land | | Live Voice WebSocket | wss://ai-live-api.original.land/live | | Agent list | https://ai.original.land/oauth/agents | | Connection status | https://ai.original.land/oauth/connections/status | | Workspace connect page | https://workspace.original.land/connect |

SDK exports

| Import | Runtime | Contents | |---|---|---| | @original-land/original-connect | any | types, PKCE utils, URL builders, discovery, error types, scopes, registerClient, extractOutputText | | @original-land/original-connect/browser | browser | createBrowserClient → login, handleCallback, getChatToken, streamResponse, listAgents, uploadFiles, connectWorkspace | | @original-land/original-connect/server | Node.js | createConfidentialClient → getAuthorizeUrl, exchangeCode, refresh, streamResponse, connectWorkspace | | @original-land/original-connect/responses | any | streamResponse, createResponse, extractOutputText (low-level) | | @original-land/original-connect/files | any | file upload utilities | | @original-land/original-connect/voice | any | speechToText, textToSpeech | | @original-land/original-connect/omni | any | streamOmniTurn, startOmniTurn, consumeOmniText, consumeOmniAudio, abortOmniConversation | | @original-land/original-connect/voice-live | browser | startVoiceSession (full-duplex WebSocket) | | @original-land/original-connect/register | any | registerClient (dynamic client registration) |

Development

pnpm install
pnpm test       # vitest
pnpm typecheck
pnpm build      # tsup → dist (ESM + CJS + d.ts)

Every client call goes through an injectable fetch, so the SDK is fully unit-testable in isolation against the documented HTTP contracts (no live issuer required).