@talk2view/sdk 0.19.0 → 0.20.1

package/docs/chat.md

@@ -109,7 +109,7 @@ take directly.
 | `systemPrompt` | none | Your own instructions, sent with every message. Changing it does not restart the conversation. |
 | `welcome` | `{}` | `heading` and `suggestions` on the empty chat. Clicking a suggestion sends it. |
 | `allowAnonymous` | `true` | Logged-out visitors chat as guests. `false` shows sign-in first — **and see the note below**. |
-| `features` | all on | `{ dictation, attachments, settings, account, threadList, skills }`. |
+| `features` | all on | `{ dictation, attachments, settings, account, threadList, skills, voice }`. `voice` is the realtime voice button beside the launcher, shown only for partners with voice enabled — see [voice.md](voice.md). |
 | `fontFamily` | inherit | Your own font stack. Default: whatever the page uses. |
 | `resetPasswordUrl` | none | Where "Forgot it?" goes. Without it the link is not shown. |
 | `footer` | none | A short line under the composer — your terms, a disclaimer. |
@@ -403,9 +403,9 @@ Stated plainly, because it is large.
 
 | | gzip |
 |---|---|
-| `@talk2view/sdk` (the core client) | 9.4 KB |
-| `@talk2view/sdk/ui` (the previous chat panel) | 39.9 KB |
-| **`@talk2view/sdk/chat`** | **361.8 KB** |
+| `@talk2view/sdk` (the core client) | 11.1 KB |
+| `@talk2view/sdk/ui` (the previous chat panel) | 40.3 KB |
+| **`@talk2view/sdk/chat`** | **372.4 KB** |
 | A whole page containing the chat and React | ~437 KB |
 
 The first three exclude React and React DOM, which your app already has. The

package/docs/voice.md

@@ -0,0 +1,88 @@
+# Voice
+
+The realtime voice agent: the end-user talks, the agent answers out loud, and
+it can use your client tools while it does. It runs the **same** tools,
+permission checks and approval card as the chat. A tool you registered for chat
+is already a voice tool, so there is no new code to write for voice.
+
+Voice is off until Talk2View turns it on for your partner account. While it is
+off, `/v1/config` returns `voice_agent_enabled: false` and nothing below shows.
+
+## In the launcher
+
+`<Talk2ViewChatLauncher>` shows a microphone beside its mark when your partner
+account has voice enabled. Press it to talk, and press it again to hang up. A
+press while the call is still connecting cancels the call. To hide the button
+even when voice is enabled:
+
+```tsx
+<Talk2ViewChatLauncher partnerKey="pk_live_…" features={{ voice: false }} />
+```
+
+Above the button, a small note shows what the call is doing: connecting,
+listening, working, or why it ended. Errors are written for the end-user, not
+the developer. For example, when every voice slot is taken, the note reads
+"Voice is busy right now. Try again shortly." When a tool needs approval, its
+card appears in the same place.
+
+## Standalone
+
+```tsx
+import { VoiceButton } from '@talk2view/sdk/chat';
+import '@talk2view/sdk/chat.css';
+
+<VoiceButton client={t2v} />;
+```
+
+| Prop | Default | |
+|---|---|---|
+| `client` | the chat's | Required outside `<Talk2ViewChat>` / the launcher. |
+| `label` | `"Talk to Talk2View"` | The button's accessible name. |
+| `earcon` | `true` | A short, locally generated sound when the microphone goes live. |
+| `className` | none | Put `dark` here for the dark palette. |
+
+A standalone `<VoiceButton>` does not check `voice_agent_enabled` itself.
+Render it only when `(await t2v.getConfig()).voice_agent_enabled` is true.
+
+## Without a UI: `t2v.voice`
+
+```ts
+await t2v.voice.start(); // resolves once the microphone is live
+t2v.voice.on('transcript', ({ role, text, final }) => { /* … */ });
+t2v.voice.on('toolCall', ({ toolName, arguments: args }) => { /* … */ });
+t2v.voice.on('approvalChange', (pending) => pending?.decide({ action: 'once' }));
+t2v.voice.on('error', ({ type, message }) => { /* … */ });
+t2v.voice.on('ended', (reason) => { /* 'stopped', 'idle', 'session_cap', … */ });
+await t2v.voice.stop(); // hangs up and releases the microphone
+```
+
+`on()` returns an unsubscribe function. Other events are `stateChange`
+(`idle | connecting | listening | ended | error`) and `agentState`
+(`working | idle`).
+
+If `start()` fails, it rejects with an error whose `.type` gives the reason:
+`voice_disabled`, `account_required`, `auth_expired`, `no_api_key`,
+`insufficient_credit`, `credit_check_unavailable`, `service_unavailable`,
+`voice_at_capacity`, `voice_ticket_invalid`, `voice_offer_invalid`,
+`upstream_error`, `transport_error` or `voice_error`. The same types arrive on
+the `error` event, which also reports what ends a call without `start()`
+rejecting: `mic_unavailable` (the microphone was refused or is missing; the
+call ends with `'error'`), `auth_expired` (the sign-in expired; the call ends with
+`'auth_expired'`) and `insufficient_credit` (credit ran out; the call ends with
+`'budget_exhausted'`). A deliberate `t2v.auth.logout()` hangs up a call as
+`'stopped'`, with no error.
+
+## Browser requirements
+
+- **HTTPS.** Browsers only give microphone access on a secure page (localhost
+  counts as secure).
+- **Microphone permission.** The browser asks on the first press. If the
+  end-user refuses, the call ends with a `mic_unavailable` error.
+- **WebRTC.** The call is a WebRTC connection to Talk2View's voice service. A
+  network that blocks WebRTC fails with `transport_error`.
+- **Content-Security-Policy.** Your bundler puts the Pipecat client libraries
+  in a separate chunk that loads from your own origin on the first press. The
+  call then connects to the voice service URL returned by the engine, so
+  `connect-src` must allow that host as well as your engine.
+
+A chat-only integration never loads the Pipecat libraries.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@talk2view/sdk",
-  "version": "0.19.0",
+  "version": "0.20.1",
   "description": "Talk2View SDK — Add AI-powered natural language control to any application",
   "type": "module",
   "main": "./dist/index.js",
@@ -82,6 +82,8 @@
     "@assistant-ui/react": "^0.15.20",
     "@assistant-ui/react-markdown": "^0.14.15",
     "@base-ui/react": "^1.8.0",
+    "@pipecat-ai/client-js": "1.13.1",
+    "@pipecat-ai/small-webrtc-transport": "1.10.8",
     "class-variance-authority": "^0.7.1",
     "clsx": "^2.1.1",
     "dompurify": "^3.4.13",