@epicbagel/saas-kit 1.11.1 → 1.12.0

package/README.md

@@ -50,8 +50,7 @@ For hidden/programmatic triggers, it uses the focused visible button as the
 desktop anchor. Hosts can pass an explicit `anchorRef` for a custom trigger.
 Without a visible anchor it opens full-screen. Nested and sibling dialogs keep
 their focus and cancellation behavior.
-The recorder runs until **Done**, shows
-**Transcribing…**, then commits the returned text once. Failed requests keep the
+The recorder runs until **Done**, finishes the last words, then commits the returned text once. Failed requests keep the
 take available for retry. Cancel stops microphone tracks, aborts the request and
 ignores a late answer. The kit never persists audio or knows provider credentials.
 
@@ -67,6 +66,21 @@ import { DictationProvider } from "@epicbagel/saas-kit/primitives/dictation-prov
 </DictationProvider>
 ```
 
+For words that appear while speaking, supply `stream` as well (or on its own).
+The host connects the provided `MediaStream` to its speech service; the kit
+captures the microphone only once. `onTranscript` replaces the live preview
+with `{ transcript, interimTranscript }`. The adapter resolves when ready and
+returns `{ finish, cancel }`: `finish()` drains the final audio and returns the
+complete transcript; `cancel()` releases the transport without saving. The
+adapter must honor `signal` during startup, streaming and finalization, and
+report unexpected disconnection through `onError`.
+
+Preview updates never invoke a control's commit callback. Done waits for
+`finish()` before saving. If streaming fails and `transcribe` is also supplied,
+the recorder shows a warning and uses its complete recording on Done. Retry
+uses that same recording. With no batch adapter, an interrupted stream is an
+error; partial words are never presented as a complete note.
+
 Set `scopeKey` from the acting user and workspace (including impersonation, if
 the app supports it). A changed scope discards the active take, stops recording
 and aborts transcription without remounting the application. Responses from the
@@ -83,8 +97,9 @@ native recognition has no audio-level source.
 failure. Choose either that result or `onTranscript` to commit; using both would
 duplicate the host action. The callback receives one complete API transcript,
 or finalized chunks with native recognition. API recording requires HTTPS (or
-localhost), microphone permission and MediaRecorder. It does not produce live
-interim text; the live voice controls are a separate feature.
+localhost), microphone permission and MediaRecorder. Live interim text requires
+the host's `stream` adapter; the batch `transcribe` adapter alone returns text
+after Done. Live voice conversation controls remain a separate feature.
 
 Without a provider, controls retain browser-native speech recognition. The
 private ui-kit's `VoiceInput` uses this same provider and capture hook, and its

package/dist/{chunk-ZZ7ICQR2.cjs → chunk-7Q4EFCUI.cjs}

@@ -3,11 +3,11 @@
 // src/hooks/use-dictation.ts
 var _react = require('react');
 var DictationContext = _react.createContext.call(void 0, null);
-function DictationProvider({ transcribe, scopeKey, children }) {
+function DictationProvider({ transcribe, stream, scopeKey, children }) {
   const scope = _react.useRef.call(void 0, { key: scopeKey, generation: 0 });
   if (scope.current.key !== scopeKey) scope.current = { key: scopeKey, generation: scope.current.generation + 1 };
   const generation = scope.current.generation;
-  const value = _react.useMemo.call(void 0, () => ({ transcribe, scope, generation }), [transcribe, generation]);
+  const value = _react.useMemo.call(void 0, () => ({ transcribe, stream, scope, generation }), [transcribe, stream, generation]);
   return _react.createElement.call(void 0, DictationContext.Provider, { value }, children);
 }
 var EMPTY = { state: "idle", transcript: "", interimTranscript: "", error: null, canRetry: false };
@@ -128,6 +128,8 @@ function releaseTracks(session) {
 }
 function dispose(session) {
   session.controller.abort();
+  _optionalChain([session, 'access', _18 => _18.streamController, 'optionalAccess', _19 => _19.abort, 'call', _20 => _20()]);
+  _optionalChain([session, 'access', _21 => _21.live, 'optionalAccess', _22 => _22.cancel, 'call', _23 => _23()]);
   if (session.nativeTimer) clearTimeout(session.nativeTimer);
   const recorder = session.recorder;
   if (recorder) {
@@ -157,9 +159,10 @@ function dispose(session) {
 }
 function useDictation({ lang = "en-US", continuous = true, onTranscript } = {}) {
   const provider = _react.useContext.call(void 0, DictationContext);
-  const transcribe = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _18 => _18.transcribe]), () => ( null));
-  const scope = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _19 => _19.scope]), () => ( null));
-  const scopeGeneration = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _20 => _20.generation]), () => ( 0));
+  const transcribe = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _24 => _24.transcribe]), () => ( null));
+  const startStream = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _25 => _25.stream]), () => ( null));
+  const scope = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _26 => _26.scope]), () => ( null));
+  const scopeGeneration = _nullishCoalesce(_optionalChain([provider, 'optionalAccess', _27 => _27.generation]), () => ( 0));
   const [snapshot, setSnapshot] = _react.useState.call(void 0, EMPTY);
   const [supported, setSupported] = _react.useState.call(void 0, false);
   const current = _react.useRef.call(void 0, null);
@@ -169,8 +172,8 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
   activeScope.current = scope;
   callback.current = onTranscript;
   _react.useEffect.call(void 0, () => {
-    setSupported(transcribe ? typeof navigator !== "undefined" && !!_optionalChain([navigator, 'access', _21 => _21.mediaDevices, 'optionalAccess', _22 => _22.getUserMedia]) && typeof MediaRecorder !== "undefined" : typeof recognitionConstructor() === "function");
-  }, [transcribe]);
+    setSupported(transcribe || startStream ? typeof navigator !== "undefined" && !!_optionalChain([navigator, 'access', _28 => _28.mediaDevices, 'optionalAccess', _29 => _29.getUserMedia]) && typeof MediaRecorder !== "undefined" : typeof recognitionConstructor() === "function");
+  }, [transcribe, startStream]);
   const cancel = _react.useCallback.call(void 0, () => {
     const session = current.current;
     current.current = null;
@@ -190,9 +193,9 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       if (session) dispose(session);
     };
   }, []);
-  const isCurrent = _react.useCallback.call(void 0, (session) => mounted.current && current.current === session && session.scope === activeScope.current && session.scopeGeneration === (_nullishCoalesce(_optionalChain([session, 'access', _23 => _23.scope, 'optionalAccess', _24 => _24.current, 'access', _25 => _25.generation]), () => ( 0))) && !session.controller.signal.aborted, []);
+  const isCurrent = _react.useCallback.call(void 0, (session) => mounted.current && current.current === session && session.scope === activeScope.current && session.scopeGeneration === (_nullishCoalesce(_optionalChain([session, 'access', _30 => _30.scope, 'optionalAccess', _31 => _31.current, 'access', _32 => _32.generation]), () => ( 0))) && !session.controller.signal.aborted, []);
   const start = _react.useCallback.call(void 0, async () => {
-    if (scope !== activeScope.current || scopeGeneration !== (_nullishCoalesce(_optionalChain([scope, 'optionalAccess', _26 => _26.current, 'access', _27 => _27.generation]), () => ( 0)))) return;
+    if (scope !== activeScope.current || scopeGeneration !== (_nullishCoalesce(_optionalChain([scope, 'optionalAccess', _33 => _33.current, 'access', _34 => _34.generation]), () => ( 0)))) return;
     if (current.current && isCurrent(current.current) && ["requesting", "recording", "transcribing"].includes(current.current.phase)) return;
     cancel();
     let resolveCapture;
@@ -200,9 +203,12 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
     const session = {
       language: lang,
       transcribe,
+      startStream,
+      liveFailed: false,
       scope,
       scopeGeneration,
       controller: new AbortController(),
+      streamController: startStream ? new AbortController() : void 0,
       phase: "requesting",
       chunks: [],
       capture: new Promise((resolve) => {
@@ -217,20 +223,20 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       transcript: "",
       interim: "",
       notified: /* @__PURE__ */ new Set(),
-      audioLevels: transcribe ? createAudioLevels() : void 0
+      audioLevels: transcribe || startStream ? createAudioLevels() : void 0
     };
     current.current = session;
     setSnapshot({ ...EMPTY, state: "requesting" });
-    if (transcribe) {
+    if (transcribe || startStream) {
       try {
-        if (!_optionalChain([navigator, 'access', _28 => _28.mediaDevices, 'optionalAccess', _29 => _29.getUserMedia]) || typeof MediaRecorder === "undefined") throw new Error("unsupported");
+        if (!_optionalChain([navigator, 'access', _35 => _35.mediaDevices, 'optionalAccess', _36 => _36.getUserMedia]) || typeof MediaRecorder === "undefined") throw new Error("unsupported");
         const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
         if (!isCurrent(session)) {
           stream.getTracks().forEach((track) => track.stop());
           return;
         }
         session.stream = stream;
-        const mimeType = MIME_TYPES.find((mime) => _optionalChain([MediaRecorder, 'access', _30 => _30.isTypeSupported, 'optionalCall', _31 => _31(mime)]));
+        const mimeType = MIME_TYPES.find((mime) => _optionalChain([MediaRecorder, 'access', _37 => _37.isTypeSupported, 'optionalCall', _38 => _38(mime)]));
         const recorder = mimeType ? new MediaRecorder(stream, { mimeType }) : new MediaRecorder(stream);
         session.recorder = recorder;
         recorder.ondataavailable = (event) => {
@@ -242,7 +248,7 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
             session.resolveCapture(null);
             return;
           }
-          const type = recorder.mimeType || _optionalChain([session, 'access', _32 => _32.chunks, 'access', _33 => _33[0], 'optionalAccess', _34 => _34.type]) || mimeType || "audio/webm";
+          const type = recorder.mimeType || _optionalChain([session, 'access', _39 => _39.chunks, 'access', _40 => _40[0], 'optionalAccess', _41 => _41.type]) || mimeType || "audio/webm";
           const blob = new Blob(session.chunks, { type });
           session.blob = blob;
           session.chunks = [];
@@ -250,16 +256,53 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
         };
         recorder.onerror = () => {
           if (!isCurrent(session)) return;
+          session.liveFailed = true;
+          _optionalChain([session, 'access', _42 => _42.live, 'optionalAccess', _43 => _43.cancel, 'call', _44 => _44()]);
           releaseTracks(session);
           session.phase = "error";
           session.resolveCapture(null);
           setSnapshot({ ...EMPTY, state: "error", error: "Recording was interrupted. Try again." });
         };
         recorder.start();
+        if (startStream) {
+          const failed = (error) => {
+            if (!isCurrent(session) || session.phase === "ready" || session.phase === "error") return;
+            session.liveFailed = true;
+            _optionalChain([session, 'access', _45 => _45.live, 'optionalAccess', _46 => _46.cancel, 'call', _47 => _47()]);
+            if (transcribe) {
+              setSnapshot((previous) => ({ ...previous, previewError: "Live text is unavailable. Choose Done to transcribe this recording." }));
+            } else {
+              releaseTracks(session);
+              if (recorder.state !== "inactive") recorder.stop();
+              session.phase = "error";
+              setSnapshot((previous) => ({ ...previous, state: "error", error: error.message || "Live transcription stopped. Try again." }));
+            }
+          };
+          try {
+            const live = await startStream(stream, {
+              language: lang,
+              signal: session.streamController.signal,
+              onTranscript: (preview) => {
+                if (!isCurrent(session) || session.liveFailed || session.phase === "error" || session.phase === "ready") return;
+                session.transcript = preview.transcript;
+                session.interim = preview.interimTranscript;
+                setSnapshot((previous) => ({ ...previous, transcript: preview.transcript, interimTranscript: preview.interimTranscript }));
+              },
+              onError: failed
+            });
+            if (!isCurrent(session) || session.liveFailed || session.phase === "error") {
+              live.cancel();
+            } else session.live = live;
+          } catch (error) {
+            failed(error instanceof Error ? error : new Error("Live transcription could not start. Try again."));
+          }
+          if (!isCurrent(session) || session.phase !== "requesting") return;
+        }
         session.phase = "recording";
         session.stopMeter = meterStream(stream, session.audioLevels, () => isCurrent(session) && session.phase === "recording");
-        setSnapshot({ ...EMPTY, state: "recording" });
+        setSnapshot((previous) => ({ ...previous, state: "recording" }));
       } catch (error) {
+        _optionalChain([session, 'access', _48 => _48.live, 'optionalAccess', _49 => _49.cancel, 'call', _50 => _50()]);
         releaseTracks(session);
         if (!isCurrent(session)) return;
         session.phase = "error";
@@ -295,7 +338,7 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
         session.transcript = final.join(" ");
         session.interim = interim.join(" ");
         setSnapshot((previous) => ({ ...previous, transcript: session.transcript, interimTranscript: session.interim }));
-        if (chunks.length) _optionalChain([callback, 'access', _35 => _35.current, 'optionalCall', _36 => _36(chunks.join(" "))]);
+        if (chunks.length) _optionalChain([callback, 'access', _51 => _51.current, 'optionalCall', _52 => _52(chunks.join(" "))]);
       };
       recognition.onend = () => {
         session.nativeEnded = true;
@@ -320,23 +363,23 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       session.phase = "error";
       setSnapshot({ ...EMPTY, state: "error", error: microphoneError(error) });
     }
-  }, [cancel, continuous, isCurrent, lang, scope, scopeGeneration, transcribe]);
+  }, [cancel, continuous, isCurrent, lang, scope, scopeGeneration, transcribe, startStream]);
   const finish = _react.useCallback.call(void 0, async (session) => {
     if (!isCurrent(session)) return null;
     releaseMeter(session);
     session.phase = "transcribing";
     setSnapshot((previous) => ({ ...previous, state: "transcribing", error: null, canRetry: false }));
-    if (!session.transcribe) {
+    if (!session.transcribe && !session.startStream) {
       if (!session.nativeEnded) {
         session.nativeTimer = setTimeout(() => {
           try {
-            _optionalChain([session, 'access', _37 => _37.recognition, 'optionalAccess', _38 => _38.abort, 'call', _39 => _39()]);
+            _optionalChain([session, 'access', _53 => _53.recognition, 'optionalAccess', _54 => _54.abort, 'call', _55 => _55()]);
           } catch (e8) {
           }
           session.resolveNativeEnd();
         }, 1e3);
         try {
-          _optionalChain([session, 'access', _40 => _40.recognition, 'optionalAccess', _41 => _41.stop, 'call', _42 => _42()]);
+          _optionalChain([session, 'access', _56 => _56.recognition, 'optionalAccess', _57 => _57.stop, 'call', _58 => _58()]);
         } catch (e9) {
           session.resolveNativeEnd();
         }
@@ -352,13 +395,29 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
     let text;
     try {
       if (!session.blob) {
-        if (_optionalChain([session, 'access', _43 => _43.recorder, 'optionalAccess', _44 => _44.state]) !== "inactive") _optionalChain([session, 'access', _45 => _45.recorder, 'optionalAccess', _46 => _46.stop, 'call', _47 => _47()]);
+        if (_optionalChain([session, 'access', _59 => _59.recorder, 'optionalAccess', _60 => _60.state]) !== "inactive") _optionalChain([session, 'access', _61 => _61.recorder, 'optionalAccess', _62 => _62.stop, 'call', _63 => _63()]);
         const blob = await session.capture;
         if (!isCurrent(session) || !blob) return null;
         session.blob = blob;
       }
-      if (!session.blob.size) throw new Error("No audio was recorded. Try again.");
-      text = (await session.transcribe(session.blob, { language: session.language, signal: session.controller.signal })).trim();
+      if (session.live && !session.liveFailed) {
+        try {
+          const final = await session.live.finish();
+          if (session.liveFailed) throw new Error("Live transcription was interrupted. Please retry this recording.");
+          text = final.trim();
+        } catch (error) {
+          session.liveFailed = true;
+          if (!session.transcribe) throw error;
+        } finally {
+          session.live.cancel();
+        }
+      }
+      if (!isCurrent(session)) return null;
+      if (text === void 0) {
+        if (!session.transcribe) throw new Error("Live transcription could not finish. Please record again.");
+        if (!session.blob.size) throw new Error("No audio was recorded. Try again.");
+        text = (await session.transcribe(session.blob, { language: session.language, signal: session.controller.signal })).trim();
+      }
       if (!isCurrent(session)) return null;
       if (!text) throw new Error("No speech was detected. Record again.");
     } catch (error) {
@@ -366,13 +425,14 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       releaseTracks(session);
       session.phase = "error";
       const message = error instanceof Error && error.message ? error.message : "Transcription failed. Try again.";
-      setSnapshot({ ...EMPTY, state: "error", error: message, canRetry: !!_optionalChain([session, 'access', _48 => _48.blob, 'optionalAccess', _49 => _49.size]) });
+      _optionalChain([session, 'access', _64 => _64.live, 'optionalAccess', _65 => _65.cancel, 'call', _66 => _66()]);
+      setSnapshot((previous) => ({ ...previous, state: "error", error: message, canRetry: !!session.transcribe && !!_optionalChain([session, 'access', _67 => _67.blob, 'optionalAccess', _68 => _68.size]) }));
       return null;
     }
     session.transcript = text;
     session.phase = "ready";
     setSnapshot({ ...EMPTY, state: "ready", transcript: text });
-    _optionalChain([callback, 'access', _50 => _50.current, 'optionalCall', _51 => _51(text)]);
+    _optionalChain([callback, 'access', _69 => _69.current, 'optionalCall', _70 => _70(text)]);
     return text;
   }, [isCurrent]);
   const stop = _react.useCallback.call(void 0, () => {
@@ -383,8 +443,13 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       return Promise.resolve(null);
     }
     if (session.phase === "requesting") {
-      cancel();
-      return Promise.resolve(null);
+      if (!session.transcribe || _optionalChain([session, 'access', _71 => _71.recorder, 'optionalAccess', _72 => _72.state]) !== "recording") {
+        cancel();
+        return Promise.resolve(null);
+      }
+      session.liveFailed = true;
+      _optionalChain([session, 'access', _73 => _73.streamController, 'optionalAccess', _74 => _74.abort, 'call', _75 => _75()]);
+      _optionalChain([session, 'access', _76 => _76.live, 'optionalAccess', _77 => _77.cancel, 'call', _78 => _78()]);
     }
     if (session.finish) return session.finish;
     if (session.phase === "error") return Promise.resolve(null);
@@ -397,7 +462,7 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       cancel();
       return Promise.resolve(null);
     }
-    if (!_optionalChain([session, 'optionalAccess', _52 => _52.transcribe]) || !_optionalChain([session, 'access', _53 => _53.blob, 'optionalAccess', _54 => _54.size])) return Promise.resolve(null);
+    if (!_optionalChain([session, 'optionalAccess', _79 => _79.transcribe]) || !_optionalChain([session, 'access', _80 => _80.blob, 'optionalAccess', _81 => _81.size])) return Promise.resolve(null);
     if (session.phase !== "error") return _nullishCoalesce(session.finish, () => ( Promise.resolve(null)));
     session.finish = finish(session);
     return session.finish;

package/dist/{chunk-EDEN7TAL.js → chunk-JKVVTFAP.js}

@@ -10,7 +10,7 @@ import {
 } from "./chunk-G6SRSWXB.js";
 import {
   useDictation
-} from "./chunk-HLAHOVEI.js";
+} from "./chunk-RPSRCB5O.js";
 
 // src/primitives/voice-dictation-button.tsx
 import {
@@ -123,7 +123,7 @@ function VoiceDictationButton({
         /* @__PURE__ */ jsx("h2", { className: cn("font-medium", anchored ? "text-sm" : "text-lg"), children: title || "Voice dictation" }),
         /* @__PURE__ */ jsxs("span", { role: "status", className: "flex items-center gap-1.5 text-xs text-muted-foreground", children: [
           busy ? /* @__PURE__ */ jsx(LoaderCircle, { "aria-hidden": "true", className: "size-3.5 animate-spin motion-reduce:animate-none" }) : /* @__PURE__ */ jsx(Mic, { "aria-hidden": "true", className: "size-3.5" }),
-          capture.state === "transcribing" ? "Transcribing\u2026" : capture.state === "requesting" ? "Waiting for microphone\u2026" : error ? "Recording not saved" : capture.state === "ready" ? "Ready" : caption
+          capture.state === "transcribing" ? "Finishing transcript\u2026" : capture.state === "requesting" ? "Starting dictation\u2026" : error ? "Recording not saved" : capture.state === "ready" ? "Ready" : caption
         ] })
       ] }),
       /* @__PURE__ */ jsxs("time", { "aria-label": "Recording duration", dateTime: `PT${elapsed}S`, className: "pt-0.5 font-mono text-xs tabular-nums text-muted-foreground", children: [
@@ -134,7 +134,7 @@ function VoiceDictationButton({
     ] }),
     /* @__PURE__ */ jsxs("div", { className: cn("flex min-h-0 flex-col gap-3", !anchored && "w-full max-w-xl flex-1 justify-center self-center py-8 text-center"), children: [
       capture.audioLevels && /* @__PURE__ */ jsx(DictationWaveform, { source: capture.audioLevels, state: capture.state, className: !anchored ? "h-20" : void 0 }),
-      /* @__PURE__ */ jsx("p", { id: descriptionId, className: "text-xs text-muted-foreground", children: capture.state === "transcribing" ? "Converting your recording to text\u2026" : capture.state === "requesting" ? "Allow microphone access to start." : error ? capture.canRetry ? "Retry this take or record again." : "Check microphone access, then try again." : "Speak, then choose Done to add the text." }),
+      /* @__PURE__ */ jsx("p", { id: descriptionId, className: "text-xs text-muted-foreground", children: capture.state === "transcribing" ? "Finishing your last words\u2026" : capture.state === "requesting" ? "Allow microphone access to start." : error ? capture.canRetry ? "Retry this take or record again." : "Try recording again." : "Speak, then choose Done to add the text." }),
       displayTranscript && /* @__PURE__ */ jsxs("p", { className: "max-h-40 overflow-y-auto text-sm leading-relaxed", children: [
         transcript,
         interim && /* @__PURE__ */ jsxs("span", { className: "text-muted-foreground", children: [
@@ -142,7 +142,8 @@ function VoiceDictationButton({
           interim
         ] })
       ] }),
-      error && /* @__PURE__ */ jsx("p", { role: "alert", className: "text-xs text-destructive", children: error })
+      error && /* @__PURE__ */ jsx("p", { role: "alert", className: "text-xs text-destructive", children: error }),
+      !error && capture.previewError && /* @__PURE__ */ jsx("p", { role: "status", className: "text-xs text-muted-foreground", children: capture.previewError })
     ] }),
     /* @__PURE__ */ jsxs("div", { className: cn("gap-2", anchored ? "flex flex-wrap items-center justify-end" : "mt-auto grid w-full max-w-xl grid-cols-2 gap-3 self-center"), children: [
       /* @__PURE__ */ jsx(Button, { ref: cancelControl, type: "button", variant: "ghost", size: "sm", className: anchored ? "min-h-[44px]" : "min-h-[48px]", onClick: handleCancel, children: "Cancel" }),

package/dist/{chunk-D47AMKT4.cjs → chunk-QWUMIAYD.cjs}

@@ -10,7 +10,7 @@ var _chunkJDEXZGIYcjs = require('./chunk-JDEXZGIY.cjs');
 var _chunkKWLSBTB2cjs = require('./chunk-KWLSBTB2.cjs');
 
 
-var _chunkZZ7ICQR2cjs = require('./chunk-ZZ7ICQR2.cjs');
+var _chunk7Q4EFCUIcjs = require('./chunk-7Q4EFCUI.cjs');
 
 // src/primitives/voice-dictation-button.tsx
 
@@ -50,7 +50,7 @@ function VoiceDictationButton({
   const isOpen = _react.useRef.call(void 0, open);
   isOpen.current = open;
   const descriptionId = _react.useId.call(void 0, );
-  const capture = _chunkZZ7ICQR2cjs.useDictation.call(void 0, { lang, continuous });
+  const capture = _chunk7Q4EFCUIcjs.useDictation.call(void 0, { lang, continuous });
   const { transcript, interimTranscript: interim, supported, error } = capture;
   const generation = _react.useRef.call(void 0, 0);
   const finishing = _react.useRef.call(void 0, false);
@@ -123,7 +123,7 @@ function VoiceDictationButton({
         /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "h2", { className: _chunkKWLSBTB2cjs.cn.call(void 0, "font-medium", anchored ? "text-sm" : "text-lg"), children: title || "Voice dictation" }),
         /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, "span", { role: "status", className: "flex items-center gap-1.5 text-xs text-muted-foreground", children: [
           busy ? /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _lucidereact.LoaderCircle, { "aria-hidden": "true", className: "size-3.5 animate-spin motion-reduce:animate-none" }) : /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _lucidereact.Mic, { "aria-hidden": "true", className: "size-3.5" }),
-          capture.state === "transcribing" ? "Transcribing\u2026" : capture.state === "requesting" ? "Waiting for microphone\u2026" : error ? "Recording not saved" : capture.state === "ready" ? "Ready" : caption
+          capture.state === "transcribing" ? "Finishing transcript\u2026" : capture.state === "requesting" ? "Starting dictation\u2026" : error ? "Recording not saved" : capture.state === "ready" ? "Ready" : caption
         ] })
       ] }),
       /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, "time", { "aria-label": "Recording duration", dateTime: `PT${elapsed}S`, className: "pt-0.5 font-mono text-xs tabular-nums text-muted-foreground", children: [
@@ -134,7 +134,7 @@ function VoiceDictationButton({
     ] }),
     /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, "div", { className: _chunkKWLSBTB2cjs.cn.call(void 0, "flex min-h-0 flex-col gap-3", !anchored && "w-full max-w-xl flex-1 justify-center self-center py-8 text-center"), children: [
       capture.audioLevels && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, DictationWaveform, { source: capture.audioLevels, state: capture.state, className: !anchored ? "h-20" : void 0 }),
-      /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "p", { id: descriptionId, className: "text-xs text-muted-foreground", children: capture.state === "transcribing" ? "Converting your recording to text\u2026" : capture.state === "requesting" ? "Allow microphone access to start." : error ? capture.canRetry ? "Retry this take or record again." : "Check microphone access, then try again." : "Speak, then choose Done to add the text." }),
+      /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "p", { id: descriptionId, className: "text-xs text-muted-foreground", children: capture.state === "transcribing" ? "Finishing your last words\u2026" : capture.state === "requesting" ? "Allow microphone access to start." : error ? capture.canRetry ? "Retry this take or record again." : "Try recording again." : "Speak, then choose Done to add the text." }),
       displayTranscript && /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, "p", { className: "max-h-40 overflow-y-auto text-sm leading-relaxed", children: [
         transcript,
         interim && /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, "span", { className: "text-muted-foreground", children: [
@@ -142,7 +142,8 @@ function VoiceDictationButton({
           interim
         ] })
       ] }),
-      error && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "p", { role: "alert", className: "text-xs text-destructive", children: error })
+      error && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "p", { role: "alert", className: "text-xs text-destructive", children: error }),
+      !error && capture.previewError && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "p", { role: "status", className: "text-xs text-muted-foreground", children: capture.previewError })
     ] }),
     /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, "div", { className: _chunkKWLSBTB2cjs.cn.call(void 0, "gap-2", anchored ? "flex flex-wrap items-center justify-end" : "mt-auto grid w-full max-w-xl grid-cols-2 gap-3 self-center"), children: [
       /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _chunkJDEXZGIYcjs.Button, { ref: cancelControl, type: "button", variant: "ghost", size: "sm", className: anchored ? "min-h-[44px]" : "min-h-[48px]", onClick: handleCancel, children: "Cancel" }),

package/dist/{chunk-HLAHOVEI.js → chunk-RPSRCB5O.js}

@@ -3,11 +3,11 @@
 // src/hooks/use-dictation.ts
 import { createContext, createElement, useCallback, useContext, useEffect, useMemo, useRef, useState } from "react";
 var DictationContext = createContext(null);
-function DictationProvider({ transcribe, scopeKey, children }) {
+function DictationProvider({ transcribe, stream, scopeKey, children }) {
   const scope = useRef({ key: scopeKey, generation: 0 });
   if (scope.current.key !== scopeKey) scope.current = { key: scopeKey, generation: scope.current.generation + 1 };
   const generation = scope.current.generation;
-  const value = useMemo(() => ({ transcribe, scope, generation }), [transcribe, generation]);
+  const value = useMemo(() => ({ transcribe, stream, scope, generation }), [transcribe, stream, generation]);
   return createElement(DictationContext.Provider, { value }, children);
 }
 var EMPTY = { state: "idle", transcript: "", interimTranscript: "", error: null, canRetry: false };
@@ -128,6 +128,8 @@ function releaseTracks(session) {
 }
 function dispose(session) {
   session.controller.abort();
+  session.streamController?.abort();
+  session.live?.cancel();
   if (session.nativeTimer) clearTimeout(session.nativeTimer);
   const recorder = session.recorder;
   if (recorder) {
@@ -158,6 +160,7 @@ function dispose(session) {
 function useDictation({ lang = "en-US", continuous = true, onTranscript } = {}) {
   const provider = useContext(DictationContext);
   const transcribe = provider?.transcribe ?? null;
+  const startStream = provider?.stream ?? null;
   const scope = provider?.scope ?? null;
   const scopeGeneration = provider?.generation ?? 0;
   const [snapshot, setSnapshot] = useState(EMPTY);
@@ -169,8 +172,8 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
   activeScope.current = scope;
   callback.current = onTranscript;
   useEffect(() => {
-    setSupported(transcribe ? typeof navigator !== "undefined" && !!navigator.mediaDevices?.getUserMedia && typeof MediaRecorder !== "undefined" : typeof recognitionConstructor() === "function");
-  }, [transcribe]);
+    setSupported(transcribe || startStream ? typeof navigator !== "undefined" && !!navigator.mediaDevices?.getUserMedia && typeof MediaRecorder !== "undefined" : typeof recognitionConstructor() === "function");
+  }, [transcribe, startStream]);
   const cancel = useCallback(() => {
     const session = current.current;
     current.current = null;
@@ -200,9 +203,12 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
     const session = {
       language: lang,
       transcribe,
+      startStream,
+      liveFailed: false,
       scope,
       scopeGeneration,
       controller: new AbortController(),
+      streamController: startStream ? new AbortController() : void 0,
       phase: "requesting",
       chunks: [],
       capture: new Promise((resolve) => {
@@ -217,11 +223,11 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       transcript: "",
       interim: "",
       notified: /* @__PURE__ */ new Set(),
-      audioLevels: transcribe ? createAudioLevels() : void 0
+      audioLevels: transcribe || startStream ? createAudioLevels() : void 0
     };
     current.current = session;
     setSnapshot({ ...EMPTY, state: "requesting" });
-    if (transcribe) {
+    if (transcribe || startStream) {
       try {
         if (!navigator.mediaDevices?.getUserMedia || typeof MediaRecorder === "undefined") throw new Error("unsupported");
         const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
@@ -250,16 +256,53 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
         };
         recorder.onerror = () => {
           if (!isCurrent(session)) return;
+          session.liveFailed = true;
+          session.live?.cancel();
           releaseTracks(session);
           session.phase = "error";
           session.resolveCapture(null);
           setSnapshot({ ...EMPTY, state: "error", error: "Recording was interrupted. Try again." });
         };
         recorder.start();
+        if (startStream) {
+          const failed = (error) => {
+            if (!isCurrent(session) || session.phase === "ready" || session.phase === "error") return;
+            session.liveFailed = true;
+            session.live?.cancel();
+            if (transcribe) {
+              setSnapshot((previous) => ({ ...previous, previewError: "Live text is unavailable. Choose Done to transcribe this recording." }));
+            } else {
+              releaseTracks(session);
+              if (recorder.state !== "inactive") recorder.stop();
+              session.phase = "error";
+              setSnapshot((previous) => ({ ...previous, state: "error", error: error.message || "Live transcription stopped. Try again." }));
+            }
+          };
+          try {
+            const live = await startStream(stream, {
+              language: lang,
+              signal: session.streamController.signal,
+              onTranscript: (preview) => {
+                if (!isCurrent(session) || session.liveFailed || session.phase === "error" || session.phase === "ready") return;
+                session.transcript = preview.transcript;
+                session.interim = preview.interimTranscript;
+                setSnapshot((previous) => ({ ...previous, transcript: preview.transcript, interimTranscript: preview.interimTranscript }));
+              },
+              onError: failed
+            });
+            if (!isCurrent(session) || session.liveFailed || session.phase === "error") {
+              live.cancel();
+            } else session.live = live;
+          } catch (error) {
+            failed(error instanceof Error ? error : new Error("Live transcription could not start. Try again."));
+          }
+          if (!isCurrent(session) || session.phase !== "requesting") return;
+        }
         session.phase = "recording";
         session.stopMeter = meterStream(stream, session.audioLevels, () => isCurrent(session) && session.phase === "recording");
-        setSnapshot({ ...EMPTY, state: "recording" });
+        setSnapshot((previous) => ({ ...previous, state: "recording" }));
       } catch (error) {
+        session.live?.cancel();
         releaseTracks(session);
         if (!isCurrent(session)) return;
         session.phase = "error";
@@ -320,13 +363,13 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       session.phase = "error";
       setSnapshot({ ...EMPTY, state: "error", error: microphoneError(error) });
     }
-  }, [cancel, continuous, isCurrent, lang, scope, scopeGeneration, transcribe]);
+  }, [cancel, continuous, isCurrent, lang, scope, scopeGeneration, transcribe, startStream]);
   const finish = useCallback(async (session) => {
     if (!isCurrent(session)) return null;
     releaseMeter(session);
     session.phase = "transcribing";
     setSnapshot((previous) => ({ ...previous, state: "transcribing", error: null, canRetry: false }));
-    if (!session.transcribe) {
+    if (!session.transcribe && !session.startStream) {
       if (!session.nativeEnded) {
         session.nativeTimer = setTimeout(() => {
           try {
@@ -357,8 +400,24 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
         if (!isCurrent(session) || !blob) return null;
         session.blob = blob;
       }
-      if (!session.blob.size) throw new Error("No audio was recorded. Try again.");
-      text = (await session.transcribe(session.blob, { language: session.language, signal: session.controller.signal })).trim();
+      if (session.live && !session.liveFailed) {
+        try {
+          const final = await session.live.finish();
+          if (session.liveFailed) throw new Error("Live transcription was interrupted. Please retry this recording.");
+          text = final.trim();
+        } catch (error) {
+          session.liveFailed = true;
+          if (!session.transcribe) throw error;
+        } finally {
+          session.live.cancel();
+        }
+      }
+      if (!isCurrent(session)) return null;
+      if (text === void 0) {
+        if (!session.transcribe) throw new Error("Live transcription could not finish. Please record again.");
+        if (!session.blob.size) throw new Error("No audio was recorded. Try again.");
+        text = (await session.transcribe(session.blob, { language: session.language, signal: session.controller.signal })).trim();
+      }
       if (!isCurrent(session)) return null;
       if (!text) throw new Error("No speech was detected. Record again.");
     } catch (error) {
@@ -366,7 +425,8 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       releaseTracks(session);
       session.phase = "error";
       const message = error instanceof Error && error.message ? error.message : "Transcription failed. Try again.";
-      setSnapshot({ ...EMPTY, state: "error", error: message, canRetry: !!session.blob?.size });
+      session.live?.cancel();
+      setSnapshot((previous) => ({ ...previous, state: "error", error: message, canRetry: !!session.transcribe && !!session.blob?.size }));
       return null;
     }
     session.transcript = text;
@@ -383,8 +443,13 @@ function useDictation({ lang = "en-US", continuous = true, onTranscript } = {})
       return Promise.resolve(null);
     }
     if (session.phase === "requesting") {
-      cancel();
-      return Promise.resolve(null);
+      if (!session.transcribe || session.recorder?.state !== "recording") {
+        cancel();
+        return Promise.resolve(null);
+      }
+      session.liveFailed = true;
+      session.streamController?.abort();
+      session.live?.cancel();
     }
     if (session.finish) return session.finish;
     if (session.phase === "error") return Promise.resolve(null);

package/dist/hooks/use-dictation.cjs

@@ -3,8 +3,8 @@
 
 
 
-var _chunkZZ7ICQR2cjs = require('../chunk-ZZ7ICQR2.cjs');
+var _chunk7Q4EFCUIcjs = require('../chunk-7Q4EFCUI.cjs');
 
 
 
-exports.DictationProvider = _chunkZZ7ICQR2cjs.DictationProvider; exports.useDictation = _chunkZZ7ICQR2cjs.useDictation;
+exports.DictationProvider = _chunk7Q4EFCUIcjs.DictationProvider; exports.useDictation = _chunk7Q4EFCUIcjs.useDictation;

package/dist/hooks/use-dictation.d.cts

@@ -9,8 +9,27 @@ interface DictationTranscribeOptions {
 }
 /** The host authenticates this request; provider credentials never belong in the kit. */
 type DictationTranscribe = (audio: Blob, options: DictationTranscribeOptions) => Promise<string>;
+interface DictationStreamOptions extends DictationTranscribeOptions {
+    /** Replace the current preview. These updates must never save a note. */
+    onTranscript: (preview: {
+        transcript: string;
+        interimTranscript: string;
+    }) => void;
+    onError: (error: Error) => void;
+}
+interface DictationStreamSession {
+    /** Drain the final audio and resolve the complete, final transcript. */
+    finish: () => Promise<string>;
+    /** Release the transport without committing any text. Must be idempotent. */
+    cancel: () => void;
+}
+/** The host owns transport/authentication; use the supplied microphone stream. */
+type DictationStream = (audio: MediaStream, options: DictationStreamOptions) => Promise<DictationStreamSession>;
 interface DictationProviderProps {
-    transcribe: DictationTranscribe;
+    /** Complete-recording transcription, also used to retry a failed stream. */
+    transcribe?: DictationTranscribe;
+    /** Optional live transcription. Preview events never invoke onTranscript. */
+    stream?: DictationStream;
     /** Change when the acting user or workspace changes to discard its active take. */
     scopeKey?: string;
     children: ReactNode;
@@ -22,12 +41,13 @@ type DictationScope = {
     };
 };
 type DictationContextValue = {
-    transcribe: DictationTranscribe;
+    transcribe?: DictationTranscribe;
+    stream?: DictationStream;
     scope: DictationScope;
     generation: number;
 };
-/** Use recorded audio + the host's API for every descendant dictation control. */
-declare function DictationProvider({ transcribe, scopeKey, children }: DictationProviderProps): React.FunctionComponentElement<React.ProviderProps<DictationContextValue | null>>;
+/** Use the host's live or recorded-audio API for every descendant control. */
+declare function DictationProvider({ transcribe, stream, scopeKey, children }: DictationProviderProps): React.FunctionComponentElement<React.ProviderProps<DictationContextValue | null>>;
 type DictationState = "idle" | "requesting" | "recording" | "transcribing" | "ready" | "error";
 interface UseDictationOptions {
     lang?: string;
@@ -42,6 +62,8 @@ interface DictationCapture {
     transcript: string;
     interimTranscript: string;
     error: string | null;
+    /** A live preview failed, but the recorded take can still be transcribed on Done. */
+    previewError?: string;
     canRetry: boolean;
     /** Live bands from the recording stream. Native speech recognition exposes no audio meter. */
     audioLevels?: {
@@ -58,4 +80,4 @@ interface DictationCapture {
 /** Shared microphone lifecycle. Without a provider, existing native dictation keeps working. */
 declare function useDictation({ lang, continuous, onTranscript }?: UseDictationOptions): DictationCapture;
 
-export { type DictationCapture, DictationProvider, type DictationProviderProps, type DictationState, type DictationTranscribe, type DictationTranscribeOptions, type UseDictationOptions, useDictation };
+export { type DictationCapture, DictationProvider, type DictationProviderProps, type DictationState, type DictationStream, type DictationStreamOptions, type DictationStreamSession, type DictationTranscribe, type DictationTranscribeOptions, type UseDictationOptions, useDictation };

package/dist/hooks/use-dictation.d.ts

@@ -9,8 +9,27 @@ interface DictationTranscribeOptions {
 }
 /** The host authenticates this request; provider credentials never belong in the kit. */
 type DictationTranscribe = (audio: Blob, options: DictationTranscribeOptions) => Promise<string>;
+interface DictationStreamOptions extends DictationTranscribeOptions {
+    /** Replace the current preview. These updates must never save a note. */
+    onTranscript: (preview: {
+        transcript: string;
+        interimTranscript: string;
+    }) => void;
+    onError: (error: Error) => void;
+}
+interface DictationStreamSession {
+    /** Drain the final audio and resolve the complete, final transcript. */
+    finish: () => Promise<string>;
+    /** Release the transport without committing any text. Must be idempotent. */
+    cancel: () => void;
+}
+/** The host owns transport/authentication; use the supplied microphone stream. */
+type DictationStream = (audio: MediaStream, options: DictationStreamOptions) => Promise<DictationStreamSession>;
 interface DictationProviderProps {
-    transcribe: DictationTranscribe;
+    /** Complete-recording transcription, also used to retry a failed stream. */
+    transcribe?: DictationTranscribe;
+    /** Optional live transcription. Preview events never invoke onTranscript. */
+    stream?: DictationStream;
     /** Change when the acting user or workspace changes to discard its active take. */
     scopeKey?: string;
     children: ReactNode;
@@ -22,12 +41,13 @@ type DictationScope = {
     };
 };
 type DictationContextValue = {
-    transcribe: DictationTranscribe;
+    transcribe?: DictationTranscribe;
+    stream?: DictationStream;
     scope: DictationScope;
     generation: number;
 };
-/** Use recorded audio + the host's API for every descendant dictation control. */
-declare function DictationProvider({ transcribe, scopeKey, children }: DictationProviderProps): React.FunctionComponentElement<React.ProviderProps<DictationContextValue | null>>;
+/** Use the host's live or recorded-audio API for every descendant control. */
+declare function DictationProvider({ transcribe, stream, scopeKey, children }: DictationProviderProps): React.FunctionComponentElement<React.ProviderProps<DictationContextValue | null>>;
 type DictationState = "idle" | "requesting" | "recording" | "transcribing" | "ready" | "error";
 interface UseDictationOptions {
     lang?: string;
@@ -42,6 +62,8 @@ interface DictationCapture {
     transcript: string;
     interimTranscript: string;
     error: string | null;
+    /** A live preview failed, but the recorded take can still be transcribed on Done. */
+    previewError?: string;
     canRetry: boolean;
     /** Live bands from the recording stream. Native speech recognition exposes no audio meter. */
     audioLevels?: {
@@ -58,4 +80,4 @@ interface DictationCapture {
 /** Shared microphone lifecycle. Without a provider, existing native dictation keeps working. */
 declare function useDictation({ lang, continuous, onTranscript }?: UseDictationOptions): DictationCapture;
 
-export { type DictationCapture, DictationProvider, type DictationProviderProps, type DictationState, type DictationTranscribe, type DictationTranscribeOptions, type UseDictationOptions, useDictation };
+export { type DictationCapture, DictationProvider, type DictationProviderProps, type DictationState, type DictationStream, type DictationStreamOptions, type DictationStreamSession, type DictationTranscribe, type DictationTranscribeOptions, type UseDictationOptions, useDictation };

package/dist/hooks/use-dictation.js

@@ -3,7 +3,7 @@
 import {
   DictationProvider,
   useDictation
-} from "../chunk-HLAHOVEI.js";
+} from "../chunk-RPSRCB5O.js";
 export {
   DictationProvider,
   useDictation

package/dist/primitives/chat.cjs

@@ -10,7 +10,7 @@ var _chunkYMI3X2YLcjs = require('../chunk-YMI3X2YL.cjs');
 var _chunk2BTDBZYNcjs = require('../chunk-2BTDBZYN.cjs');
 
 
-var _chunkD47AMKT4cjs = require('../chunk-D47AMKT4.cjs');
+var _chunkQWUMIAYDcjs = require('../chunk-QWUMIAYD.cjs');
 
 
 
@@ -86,7 +86,7 @@ require('../chunk-2BSFYU22.cjs');
 
 
 var _chunk4LXN6GLZcjs = require('../chunk-4LXN6GLZ.cjs');
-require('../chunk-ZZ7ICQR2.cjs');
+require('../chunk-7Q4EFCUI.cjs');
 
 
 var _chunkR35FT53Gcjs = require('../chunk-R35FT53G.cjs');
@@ -402,8 +402,9 @@ function ChatInput({
   const [editing, setEditing] = _react.useState.call(void 0, false);
   const focusVersion = _react.useRef.call(void 0, 0);
   const isDisabled = pending || disabled;
+  const hasText = Boolean(value.trim());
   const attachmentsReady = attachments.length > 0 && attachments.every((item) => item.state !== "uploading");
-  const canSubmit = !isDisabled && (Boolean(value.trim()) || attachmentsReady);
+  const canSubmit = !isDisabled && (hasText || attachmentsReady);
   const liveVoiceActive = voiceSession && voiceSession.state !== "disconnected" && voiceSession.state !== "error";
   const submit = () => {
     const trimmed = value.trim();
@@ -541,7 +542,7 @@ function ChatInput({
                       ) : null,
                       (!helperText || compact) && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, "div", { className: "flex-1" }),
                       voice && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, 
-                        _chunkD47AMKT4cjs.VoiceDictationButton,
+                        _chunkQWUMIAYDcjs.VoiceDictationButton,
                         {
                           size: "icon",
                           title: voiceTitle,
@@ -555,8 +556,7 @@ function ChatInput({
                           )
                         }
                       ),
-                      voiceSession && !liveVoiceActive && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _chunkYMI3X2YLcjs.VoiceSessionControls, { ...voiceSession, disabled: voiceSession.disabled || isDisabled }),
-                      submitLabel ? /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, 
+                      voiceSession && !liveVoiceActive && !hasText && attachments.length === 0 ? /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _chunkYMI3X2YLcjs.VoiceSessionControls, { ...voiceSession, disabled: voiceSession.disabled || isDisabled }) : submitLabel ? /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, 
                         _chunkJDEXZGIYcjs.Button,
                         {
                           type: "submit",
@@ -572,11 +572,15 @@ function ChatInput({
                         _chunkJDEXZGIYcjs.Button,
                         {
                           type: "submit",
-                          size: spacious && !compact ? "icon-lg" : "icon",
+                          size: !voiceSession && spacious && !compact ? "icon-lg" : "icon",
+                          shape: voiceSession ? "circle" : void 0,
                           disabled: !canSubmit,
                           "aria-label": "Send",
                           "data-chat-action": "submit",
-                          className: _chunkKWLSBTB2cjs.cn.call(void 0, spacious && "rounded-xl", compactOnBlur && "min-h-[44px] min-w-[44px] shrink-0"),
+                          className: _chunkKWLSBTB2cjs.cn.call(void 0, 
+                            spacious && !voiceSession && "rounded-xl",
+                            (voiceSession || compactOnBlur) && "min-h-[44px] min-w-[44px] shrink-0"
+                          ),
                           children: /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _lucidereact.ArrowUp, { className: _chunkKWLSBTB2cjs.cn.call(void 0, spacious && "h-5 w-5") })
                         }
                       )
@@ -616,7 +620,7 @@ function VoiceHero({
     voiceSession ? /* @__PURE__ */ _jsxruntime.jsxs.call(void 0, _jsxruntime.Fragment, { children: [
       /* @__PURE__ */ _jsxruntime.jsx.call(void 0, _chunkYMI3X2YLcjs.VoiceSessionControls, { ...voiceSession, size: "hero" }),
       voice && /* @__PURE__ */ _jsxruntime.jsx.call(void 0, 
-        _chunkD47AMKT4cjs.VoiceDictationButton,
+        _chunkQWUMIAYDcjs.VoiceDictationButton,
         {
           size: "icon",
           title: voiceTitle,
@@ -626,7 +630,7 @@ function VoiceHero({
         }
       )
     ] }) : /* @__PURE__ */ _jsxruntime.jsx.call(void 0, 
-      _chunkD47AMKT4cjs.VoiceDictationButton,
+      _chunkQWUMIAYDcjs.VoiceDictationButton,
       {
         size: "lg",
         title: voiceTitle,

package/dist/primitives/chat.d.cts

@@ -154,7 +154,11 @@ interface ChatInputProps {
     uploadingLabel?: string;
     /** Show the inline voice dictation button. Defaults to true. */
     voice?: boolean;
-    /** A separate live voice control alongside dictation. The host owns its transport and lifecycle. */
+    /**
+     * Live voice controls, separate from dictation. With no text or staged files,
+     * the round voice action replaces Send; an active call keeps its controls
+     * above the input. The host owns the transport and lifecycle.
+     */
     voiceSession?: VoiceSessionControlsProps;
     /** Title shown inside the voice overlay. */
     voiceTitle?: string;
@@ -164,6 +168,7 @@ interface ChatInputProps {
      * Label for the submit button. When set, the button renders as a labelled
      * primary action (`{label} →`) instead of the default icon-only send — use
      * it to name the outcome, e.g. "Propose 3 times", "Create link", "Send".
+     * An empty composer with live voice still shows the voice action first.
      */
     submitLabel?: ReactNode;
     /** Auto-focus the textarea on mount. */
@@ -200,9 +205,10 @@ interface ChatInputProps {
 /**
  * Compact text input: a form around one bordered input group, with any
  * staged files on top, a textarea that grows with its text, then an action
- * bar (the file menu and voice mic on the left, the submit button on the
- * right). Always text-first; voice is a secondary affordance. Enter sends;
- * Shift+Enter newlines, and Enter that commits an IME candidate never sends.
+ * bar ending in one primary action. When live voice is available, an empty
+ * composer shows its round start action; text or staged files replace it
+ * with Send. Dictation remains a separate mic. Enter sends; Shift+Enter
+ * newlines, and Enter that commits an IME candidate never sends or starts a call.
  *
  * The submit button is icon-only by default; pass `submitLabel` to name the
  * outcome ("Propose 3 times", "Create link") — it then renders as a labelled
@@ -294,7 +300,8 @@ interface ChatProps {
      * Label for the submit button (empty-state input). When set, the button
      * names the outcome (`{label} →`) instead of the icon-only send — e.g.
      * "Propose 3 times". Once the conversation is active the input reverts to
-     * the compact icon send.
+     * the compact icon send. With live voice available, an empty composer shows
+     * the voice action until text or files are staged.
      */
     submitLabel?: ReactNode;
     /** Files currently staged in the composer. */
@@ -357,7 +364,7 @@ interface ChatProps {
     fileDropDescription?: ReactNode;
     /** Enable voice dictation. Defaults to true. */
     voice?: boolean;
-    /** Controlled live voice UI, used instead of browser dictation when provided. */
+    /** Controlled live voice UI; its empty-input action is separate from the dictation mic. */
     voiceSession?: VoiceSessionControlsProps;
     /** Title shown inside the voice overlay. */
     voiceTitle?: string;

package/dist/primitives/chat.d.ts

@@ -154,7 +154,11 @@ interface ChatInputProps {
     uploadingLabel?: string;
     /** Show the inline voice dictation button. Defaults to true. */
     voice?: boolean;
-    /** A separate live voice control alongside dictation. The host owns its transport and lifecycle. */
+    /**
+     * Live voice controls, separate from dictation. With no text or staged files,
+     * the round voice action replaces Send; an active call keeps its controls
+     * above the input. The host owns the transport and lifecycle.
+     */
     voiceSession?: VoiceSessionControlsProps;
     /** Title shown inside the voice overlay. */
     voiceTitle?: string;
@@ -164,6 +168,7 @@ interface ChatInputProps {
      * Label for the submit button. When set, the button renders as a labelled
      * primary action (`{label} →`) instead of the default icon-only send — use
      * it to name the outcome, e.g. "Propose 3 times", "Create link", "Send".
+     * An empty composer with live voice still shows the voice action first.
      */
     submitLabel?: ReactNode;
     /** Auto-focus the textarea on mount. */
@@ -200,9 +205,10 @@ interface ChatInputProps {
 /**
  * Compact text input: a form around one bordered input group, with any
  * staged files on top, a textarea that grows with its text, then an action
- * bar (the file menu and voice mic on the left, the submit button on the
- * right). Always text-first; voice is a secondary affordance. Enter sends;
- * Shift+Enter newlines, and Enter that commits an IME candidate never sends.
+ * bar ending in one primary action. When live voice is available, an empty
+ * composer shows its round start action; text or staged files replace it
+ * with Send. Dictation remains a separate mic. Enter sends; Shift+Enter
+ * newlines, and Enter that commits an IME candidate never sends or starts a call.
  *
  * The submit button is icon-only by default; pass `submitLabel` to name the
  * outcome ("Propose 3 times", "Create link") — it then renders as a labelled
@@ -294,7 +300,8 @@ interface ChatProps {
      * Label for the submit button (empty-state input). When set, the button
      * names the outcome (`{label} →`) instead of the icon-only send — e.g.
      * "Propose 3 times". Once the conversation is active the input reverts to
-     * the compact icon send.
+     * the compact icon send. With live voice available, an empty composer shows
+     * the voice action until text or files are staged.
      */
     submitLabel?: ReactNode;
     /** Files currently staged in the composer. */
@@ -357,7 +364,7 @@ interface ChatProps {
     fileDropDescription?: ReactNode;
     /** Enable voice dictation. Defaults to true. */
     voice?: boolean;
-    /** Controlled live voice UI, used instead of browser dictation when provided. */
+    /** Controlled live voice UI; its empty-input action is separate from the dictation mic. */
     voiceSession?: VoiceSessionControlsProps;
     /** Title shown inside the voice overlay. */
     voiceTitle?: string;

package/dist/primitives/chat.js

@@ -10,7 +10,7 @@ import {
 } from "../chunk-Z45YKVAM.js";
 import {
   VoiceDictationButton
-} from "../chunk-EDEN7TAL.js";
+} from "../chunk-JKVVTFAP.js";
 import {
   Suggestion,
   Suggestions
@@ -86,7 +86,7 @@ import "../chunk-GJ7S3L6R.js";
 import {
   useCopyToClipboard
 } from "../chunk-ZWZAAGQF.js";
-import "../chunk-HLAHOVEI.js";
+import "../chunk-RPSRCB5O.js";
 import {
   useIsMobile
 } from "../chunk-PIU5SSUV.js";
@@ -402,8 +402,9 @@ function ChatInput({
   const [editing, setEditing] = useState(false);
   const focusVersion = useRef(0);
   const isDisabled = pending || disabled;
+  const hasText = Boolean(value.trim());
   const attachmentsReady = attachments.length > 0 && attachments.every((item) => item.state !== "uploading");
-  const canSubmit = !isDisabled && (Boolean(value.trim()) || attachmentsReady);
+  const canSubmit = !isDisabled && (hasText || attachmentsReady);
   const liveVoiceActive = voiceSession && voiceSession.state !== "disconnected" && voiceSession.state !== "error";
   const submit = () => {
     const trimmed = value.trim();
@@ -555,8 +556,7 @@ function ChatInput({
                           )
                         }
                       ),
-                      voiceSession && !liveVoiceActive && /* @__PURE__ */ jsx(VoiceSessionControls, { ...voiceSession, disabled: voiceSession.disabled || isDisabled }),
-                      submitLabel ? /* @__PURE__ */ jsxs(
+                      voiceSession && !liveVoiceActive && !hasText && attachments.length === 0 ? /* @__PURE__ */ jsx(VoiceSessionControls, { ...voiceSession, disabled: voiceSession.disabled || isDisabled }) : submitLabel ? /* @__PURE__ */ jsxs(
                         Button,
                         {
                           type: "submit",
@@ -572,11 +572,15 @@ function ChatInput({
                         Button,
                         {
                           type: "submit",
-                          size: spacious && !compact ? "icon-lg" : "icon",
+                          size: !voiceSession && spacious && !compact ? "icon-lg" : "icon",
+                          shape: voiceSession ? "circle" : void 0,
                           disabled: !canSubmit,
                           "aria-label": "Send",
                           "data-chat-action": "submit",
-                          className: cn(spacious && "rounded-xl", compactOnBlur && "min-h-[44px] min-w-[44px] shrink-0"),
+                          className: cn(
+                            spacious && !voiceSession && "rounded-xl",
+                            (voiceSession || compactOnBlur) && "min-h-[44px] min-w-[44px] shrink-0"
+                          ),
                           children: /* @__PURE__ */ jsx(ArrowUp, { className: cn(spacious && "h-5 w-5") })
                         }
                       )

package/dist/primitives/dictation-provider.cjs

@@ -2,7 +2,7 @@
 "use client";
 
 
-var _chunkZZ7ICQR2cjs = require('../chunk-ZZ7ICQR2.cjs');
+var _chunk7Q4EFCUIcjs = require('../chunk-7Q4EFCUI.cjs');
 
 
-exports.DictationProvider = _chunkZZ7ICQR2cjs.DictationProvider;
+exports.DictationProvider = _chunk7Q4EFCUIcjs.DictationProvider;

package/dist/primitives/dictation-provider.d.cts

@@ -1,2 +1,2 @@
-export { DictationProvider, DictationProviderProps, DictationTranscribe, DictationTranscribeOptions } from '../hooks/use-dictation.cjs';
+export { DictationProvider, DictationProviderProps, DictationStream, DictationStreamOptions, DictationStreamSession, DictationTranscribe, DictationTranscribeOptions } from '../hooks/use-dictation.cjs';
 import 'react';

package/dist/primitives/dictation-provider.d.ts

@@ -1,2 +1,2 @@
-export { DictationProvider, DictationProviderProps, DictationTranscribe, DictationTranscribeOptions } from '../hooks/use-dictation.js';
+export { DictationProvider, DictationProviderProps, DictationStream, DictationStreamOptions, DictationStreamSession, DictationTranscribe, DictationTranscribeOptions } from '../hooks/use-dictation.js';
 import 'react';

package/dist/primitives/dictation-provider.js

@@ -2,7 +2,7 @@
 "use client";
 import {
   DictationProvider
-} from "../chunk-HLAHOVEI.js";
+} from "../chunk-RPSRCB5O.js";
 export {
   DictationProvider
 };

package/dist/primitives/voice-dictation-button.cjs

@@ -2,12 +2,12 @@
 "use client";
 
 
-var _chunkD47AMKT4cjs = require('../chunk-D47AMKT4.cjs');
+var _chunkQWUMIAYDcjs = require('../chunk-QWUMIAYD.cjs');
 require('../chunk-NKWHDI4C.cjs');
 require('../chunk-JDEXZGIY.cjs');
 require('../chunk-KWLSBTB2.cjs');
 require('../chunk-2BSFYU22.cjs');
-require('../chunk-ZZ7ICQR2.cjs');
+require('../chunk-7Q4EFCUI.cjs');
 
 
-exports.VoiceDictationButton = _chunkD47AMKT4cjs.VoiceDictationButton;
+exports.VoiceDictationButton = _chunkQWUMIAYDcjs.VoiceDictationButton;

package/dist/primitives/voice-dictation-button.js

@@ -2,12 +2,12 @@
 "use client";
 import {
   VoiceDictationButton
-} from "../chunk-EDEN7TAL.js";
+} from "../chunk-JKVVTFAP.js";
 import "../chunk-5OF6D3GH.js";
 import "../chunk-W326GKML.js";
 import "../chunk-G6SRSWXB.js";
 import "../chunk-GJ7S3L6R.js";
-import "../chunk-HLAHOVEI.js";
+import "../chunk-RPSRCB5O.js";
 export {
   VoiceDictationButton
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@epicbagel/saas-kit",
-  "version": "1.11.1",
+  "version": "1.12.0",
   "description": "Accessible React primitives, layout components and design tokens for SaaS apps, built on Radix UI and Tailwind CSS v4.",
   "license": "MIT",
   "keywords": [