feat: workspaces, agentic action tools, local Whisper STT, vec recall

- Projects/workspaces: documents grouped into projects; chat RAG scopes to the
  active project. Switcher in the Documents page.
- Agentic action tools: web_search, fetch_url, and remember (first write tool),
  allowlist-gated per playbook.
- Local Whisper STT (faster-whisper, no torch): on-device dictation replacing
  the browser Web Speech API. POST /stt + GET /stt/status; browser fallback.
- Vector index extended to conversation recall (message_vectors), with the
  brute-force cosine scan kept as the fallback.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

interface/web/src/Chatbot.jsx

@@ -19,23 +19,64 @@ export function Chatbot({ conversationId, setConversationId, onConversationChang
   const [editingIdx, setEditingIdx] = useState(null); // user message being edited
   const [editText, setEditText] = useState("");
   const [listening, setListening] = useState(false); // mic dictation active
+  const [transcribing, setTranscribing] = useState(false); // local STT running
+  const [sttLocal, setSttLocal] = useState(false); // backend Whisper available
   const [speakingIdx, setSpeakingIdx] = useState(null); // message being read aloud
   const recognitionRef = useRef(null);
+  const mediaRecRef = useRef(null);
 
   // Web Speech API — browser-native, no backend/model. Absent on unsupported browsers.
   const SpeechRec = typeof window !== "undefined" && (window.SpeechRecognition || window.webkitSpeechRecognition);
   const ttsSupported = typeof window !== "undefined" && "speechSynthesis" in window;
+  const canRecord = typeof navigator !== "undefined" && navigator.mediaDevices && window.MediaRecorder;
 
-  const toggleMic = () => {
-    if (!SpeechRec) return;
-    if (listening) { recognitionRef.current?.stop(); return; }
+  // Prefer local Whisper (on-device) over browser speech (Chrome routes audio to Google).
+  useEffect(() => {
+    fetch(`${API_BASE}/stt/status`).then(r => r.ok ? r.json() : null)
+      .then(d => setSttLocal(Boolean(d && d.available))).catch(() => {});
+  }, []);
+
+  const _appendTranscript = (text) => {
+    if (text) setInput(prev => (prev ? prev + " " : "") + text);
+  };
+
+  // Local path: record audio, POST to /stt (faster-whisper transcribes on-device).
+  const startLocalDictation = async () => {
+    try {
+      const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
+      const rec = new MediaRecorder(stream);
+      const chunks = [];
+      rec.ondataavailable = e => e.data.size && chunks.push(e.data);
+      rec.onstop = async () => {
+        stream.getTracks().forEach(t => t.stop());
+        setListening(false);
+        setTranscribing(true);
+        try {
+          const b64 = await new Promise((res) => {
+            const fr = new FileReader();
+            fr.onload = () => res(String(fr.result).split(",")[1]);
+            fr.readAsDataURL(new Blob(chunks, { type: rec.mimeType }));
+          });
+          const r = await fetch(`${API_BASE}/stt`, {
+            method: "POST", headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({ audio: b64 }),
+          });
+          if (r.ok) _appendTranscript((await r.json()).text);
+        } catch { /* ignore */ } finally {
+          setTranscribing(false);
+        }
+      };
+      mediaRecRef.current = rec;
+      setListening(true);
+      rec.start();
+    } catch { setListening(false); }
+  };
+
+  const startBrowserDictation = () => {
     const rec = new SpeechRec();
     rec.lang = "en-US";
     rec.interimResults = false;
-    rec.onresult = (e) => {
-      const text = Array.from(e.results).map(r => r[0].transcript).join(" ").trim();
-      if (text) setInput(prev => (prev ? prev + " " : "") + text);
-    };
+    rec.onresult = (e) => _appendTranscript(Array.from(e.results).map(r => r[0].transcript).join(" ").trim());
     rec.onend = () => setListening(false);
     rec.onerror = () => setListening(false);
     recognitionRef.current = rec;
@@ -43,6 +84,17 @@ export function Chatbot({ conversationId, setConversationId, onConversationChang
     rec.start();
   };
 
+  const toggleMic = () => {
+    if (transcribing) return;
+    if (listening) {
+      if (sttLocal && mediaRecRef.current) mediaRecRef.current.stop();
+      else recognitionRef.current?.stop();
+      return;
+    }
+    if (sttLocal && canRecord) startLocalDictation();
+    else if (SpeechRec) startBrowserDictation();
+  };
+
   // Read a reply aloud (light markdown strip so symbols aren't spoken).
   const speak = (idx, text) => {
     if (!ttsSupported) return;
@@ -617,10 +669,11 @@ export function Chatbot({ conversationId, setConversationId, onConversationChang
             📎
             <input type="file" accept="image/*" multiple onChange={onImagePick} disabled={loading} style={{ display: "none" }} />
           </label>
-          {SpeechRec && (
-            <button onClick={toggleMic} title={listening ? "Stop dictation" : "Dictate (speech to text)"}
-              style={{ padding: "0.9rem", background: listening ? "#3a1a1a" : "#222", border: "1px solid " + (listening ? "#c0392b" : "#333"), borderRadius: "10px", cursor: "pointer", color: listening ? "#ff8a80" : "#ccc" }}>
-              {listening ? "🔴" : "🎤"}
+          {((sttLocal && canRecord) || SpeechRec) && (
+            <button onClick={toggleMic}
+              title={transcribing ? "Transcribing…" : listening ? "Stop dictation" : sttLocal ? "Dictate (on-device Whisper)" : "Dictate (browser speech)"}
+              style={{ padding: "0.9rem", background: listening ? "#3a1a1a" : "#222", border: "1px solid " + (listening ? "#c0392b" : "#333"), borderRadius: "10px", cursor: transcribing ? "default" : "pointer", color: listening ? "#ff8a80" : "#ccc" }}>
+              {transcribing ? "⏳" : listening ? "🔴" : "🎤"}
             </button>
           )}
           <textarea