@kodaris/krubble-app-components 1.0.99 → 1.0.101

package/dist/krubble-app.umd.js

@@ -5629,6 +5629,24 @@
            * in place; call requestUpdate() after changing it to re-render.
            */
           this._globalSkillPending = new Set();
+          /**
+           * The live Web Speech recognizer while dictating, else null. Not reactive —
+           * `_isDictating` drives the UI; this is just the handle used to stop it.
+           */
+          this._speechRecognition = null;
+          /**
+           * The composer text as it stood when dictation started. Each `result` event
+           * re-reports the whole utterance (final + interim), so the value is rebuilt as
+           * this base + the transcript rather than appended to — appending would repeat
+           * the interim text on every event.
+           */
+          this._speechBaseValue = '';
+          /**
+           * Transcript of utterances the recognizer has marked final this session.
+           * Held separately because `results` is pruned in continuous mode on some
+           * browsers; without this, earlier sentences would vanish mid-dictation.
+           */
+          this._speechFinalTranscript = '';
           // --- @property (public API) ---
           /**
            * Callback function for sending messages. Receives the user's message and
@@ -5778,6 +5796,8 @@
           this._skillBrowseSource = 'org';
           /** Search text in the browse dialog, routed to the active tab's search. */
           this._skillBrowseQuery = '';
+          /** True while the mic is live and speech is being transcribed into the composer. */
+          this._isDictating = false;
           this._handleMouseMove = (e) => {
               if (this._isDragging) {
                   let newLeft = this._dragStartLeft + (e.clientX - this._dragStartX);
@@ -5930,6 +5950,14 @@
           }
           document.removeEventListener('mousemove', this._handleMouseMove);
           document.removeEventListener('mouseup', this._handleMouseUp);
+          // Release the mic if the chatbot is torn down mid-dictation — otherwise the
+          // recognizer keeps the browser's recording indicator lit with nothing to
+          // transcribe into. abort() (not stop()) since the discarded text has nowhere to go.
+          if (this._speechRecognition) {
+              this._speechRecognition.abort();
+              this._speechRecognition = null;
+              this._isDictating = false;
+          }
           // Release any object URLs still held by staged attachments.
           this._pendingConversationFiles.forEach((p) => {
               if (p.previewUrl) {
@@ -6536,6 +6564,92 @@
           this._skillMenuOpened = true;
           this._searchInvocableSkills('');
       }
+      /**
+       * The mic button toggles dictation. Speech is transcribed straight into the
+       * composer as editable text — the user still reviews and presses send, so a
+       * misheard word is fixable and nothing is sent by voice alone.
+       *
+       * Recognition runs entirely in the browser's Web Speech API (Chrome/Edge send
+       * audio to Google's service, Safari to Apple's). No audio reaches Kodaris, and
+       * there's no backend call or cost on our side.
+       */
+      _handleMicClick() {
+          // Second click stops. `stop()` flushes the pending utterance as a final
+          // result, so the last words aren't lost; onend clears the state.
+          if (this._speechRecognition) {
+              this._speechRecognition.stop();
+              return;
+          }
+          // Prefixed in Chrome/Edge/Safari; the unprefixed name is checked first for
+          // whenever they drop the prefix. The mic button isn't rendered without one of
+          // these, so reaching here unsupported means a mid-session change — bail quietly.
+          const SpeechRecognitionCtor = window.SpeechRecognition || window.webkitSpeechRecognition;
+          if (!SpeechRecognitionCtor) {
+              return;
+          }
+          const recognition = new SpeechRecognitionCtor();
+          // Keep listening through natural pauses — a chat message is often several
+          // sentences, and the default stops at the first one.
+          recognition.continuous = true;
+          // Show words as they're recognized rather than only at end of utterance.
+          recognition.interimResults = true;
+          // `lang` is deliberately left unset so the recognizer follows the page/OS
+          // language, which is correct for every consumer of this component.
+          // Dictation ADDS to whatever is already typed rather than replacing it, so
+          // the user can type half a message and speak the rest. The separating space
+          // is part of the base so the transcript never runs into the existing word.
+          this._speechBaseValue = this._inputValue;
+          if (this._speechBaseValue && !this._speechBaseValue.endsWith(' ')) {
+              this._speechBaseValue += ' ';
+          }
+          this._speechFinalTranscript = '';
+          recognition.onresult = (e) => {
+              // Everything from resultIndex on is new since the last event: finalized
+              // pieces move to _speechFinalTranscript permanently, while non-final ones
+              // are rebuilt each event because the browser revises its guess as it hears
+              // more. Rebuilding the whole value (base + final + interim) is what keeps
+              // those revisions from stacking up as duplicated text.
+              let interim = '';
+              for (let i = e.resultIndex; i < e.results.length; i++) {
+                  if (e.results[i].isFinal) {
+                      this._speechFinalTranscript += e.results[i][0].transcript;
+                  }
+                  else {
+                      interim += e.results[i][0].transcript;
+                  }
+              }
+              this._inputValue = this._speechBaseValue + this._speechFinalTranscript + interim;
+              this._saveState();
+              // Grow the textarea as dictated text wraps — the same clamp the typing path
+              // uses, since @input doesn't fire for programmatic value changes.
+              if (this._inputEl) {
+                  this._inputEl.style.height = 'auto';
+                  this._inputEl.style.height = Math.min(this._inputEl.scrollHeight, 120) + 'px';
+              }
+          };
+          recognition.onerror = (e) => {
+              // Denied mic permission is the one error worth explaining — the user has to
+              // fix it in browser settings, and the button would otherwise look broken.
+              if (e.error === 'not-allowed' || e.error === 'service-not-allowed') {
+                  this._error = 'Microphone access is blocked. Allow it in your browser settings to use voice input.';
+                  return;
+              }
+              // `no-speech` (silence) and `aborted` (user stopped) are normal; anything
+              // else is a transient service hiccup. onend still runs, so the mic resets.
+          };
+          // Fires on stop(), on error, and when the browser ends recognition on its own
+          // (e.g. a long silence) — the single place the live state is torn down.
+          recognition.onend = () => {
+              this._speechRecognition = null;
+              this._isDictating = false;
+          };
+          // start() is what triggers the browser's mic permission prompt.
+          this._speechRecognition = recognition;
+          this._isDictating = true;
+          this._error = '';
+          recognition.start();
+          this._inputEl?.focus();
+      }
       /** Insert `/code ` into the composer and focus it, ready for the message. */
       _handleSkillSelect(skill) {
           // A global skill must be installed before it can be chosen — inserting /code
@@ -6806,6 +6920,13 @@
                   previewUrl: p.previewUrl,
               }));
           }
+          // The message is on its way, so stop the mic if it's still live — otherwise
+          // it would keep transcribing into the now-cleared composer. abort() rather
+          // than stop() because a trailing pending utterance belongs to the message
+          // just sent; flushing it would drop stray words into the next one.
+          if (this._speechRecognition) {
+              this._speechRecognition.abort();
+          }
           this.messages = [...this.messages, userMessage];
           this._inputValue = '';
           this._saveState();
@@ -7878,6 +7999,31 @@
               @keydown=${this._handleKeyDown}
               @paste=${this._handlePaste}
             ></textarea>
+            ${window.SpeechRecognition || window.webkitSpeechRecognition ? b `
+              <button
+                class=${e({
+            mic: true,
+            'mic--recording': this._isDictating,
+        })}
+                title=${this._isDictating ? 'Stop dictation' : 'Dictate a message'}
+                @click=${this._handleMicClick}
+              >
+                <svg
+                  xmlns="http://www.w3.org/2000/svg"
+                  viewBox="0 0 24 24"
+                  fill="none"
+                  stroke="currentColor"
+                  stroke-width="1.75"
+                  stroke-linecap="round"
+                  stroke-linejoin="round"
+                  class="mic-icon"
+                >
+                  <path d="M12 15a3 3 0 0 0 3-3V6a3 3 0 0 0-6 0v6a3 3 0 0 0 3 3Z"/>
+                  <path d="M19 11a7 7 0 0 1-14 0"/>
+                  <path d="M12 18v3"/>
+                </svg>
+              </button>
+            ` : A}
             ${this._isStreaming ? b `
               <button
                 class="stop"
@@ -8721,6 +8867,51 @@
       color: #b0b0b0;
     }
 
+    .mic {
+      width: 34px;
+      height: 34px;
+      border: none;
+      background: transparent;
+      color: var(--kr-chatbot-text);
+      border-radius: 50%;
+      cursor: pointer;
+      display: flex;
+      align-items: center;
+      justify-content: center;
+      flex-shrink: 0;
+      transition: background 0.15s, color 0.15s;
+    }
+
+    .mic:hover {
+      background: #f8f9fc;
+    }
+
+    /* Live dictation: red mic on a tinted circle, pulsing so it's clear the
+       microphone is still open even when the user has stopped speaking. */
+    .mic--recording {
+      color: var(--kr-chatbot-error-text);
+      background: var(--kr-chatbot-error-bg);
+      animation: chatbot-mic-pulse 1.6s ease-in-out infinite;
+    }
+
+    .mic--recording:hover {
+      background: var(--kr-chatbot-error-bg);
+    }
+
+    .mic-icon {
+      width: 18px;
+      height: 18px;
+    }
+
+    @keyframes chatbot-mic-pulse {
+      0%, 100% {
+        opacity: 1;
+      }
+      50% {
+        opacity: 0.55;
+      }
+    }
+
     .send {
       width: 34px;
       height: 34px;
@@ -9348,6 +9539,9 @@
   __decorate([
       r()
   ], exports.Chatbot.prototype, "_skillBrowseQuery", void 0);
+  __decorate([
+      r()
+  ], exports.Chatbot.prototype, "_isDictating", void 0);
   __decorate([
       e$3('.panel')
   ], exports.Chatbot.prototype, "_panelEl", void 0);