diff --git a/frontend/web/src/lib/api.ts b/frontend/web/src/lib/api.ts
index c7b7d78..bbb1aea 100644
--- a/frontend/web/src/lib/api.ts
+++ b/frontend/web/src/lib/api.ts
@@ -248,15 +248,24 @@ export function chatSync(
  *  `image` (a base64 data: URI or URL) is attached for the vision path; the
  *  gateway's chat body currently has no image field, so an unconfigured backend
  *  simply ignores it. `deviceId` mirrors the client's per-browser id onto the
- *  persisted message (design/16 §4.5). */
+ *  persisted message (design/16 §4.5). `mode` is the behavior mode this turn is
+ *  sent in (design/01 §3); omitted when empty so the backend falls back to chat. */
 export function chatStream(
   sessionId: string,
   text: string,
-  image?: string
+  image?: string,
+  mode?: Mode
 ): Promise<{ stream: true }> {
   return request<{ stream: true }>("/chat", {
     method: "POST",
-    body: JSON.stringify({ sessionId, text, image, stream: true, deviceId: getClientId() }),
+    body: JSON.stringify({
+      sessionId,
+      text,
+      image,
+      stream: true,
+      deviceId: getClientId(),
+      ...(mode ? { mode } : {}),
+    }),
   });
 }
 
diff --git a/frontend/web/src/lib/useChat.ts b/frontend/web/src/lib/useChat.ts
index f32220e..c74e3c5 100644
--- a/frontend/web/src/lib/useChat.ts
+++ b/frontend/web/src/lib/useChat.ts
@@ -8,13 +8,13 @@ import { now } from "./util";
 
 /** Drives one user turn: push the user bubble, blanket a streaming assistant
  *  bubble, trigger the backend loop over REST (`stream:true`), and let the WS
- *  stream render into that bubble. Falls back to the WS `chat` channel when the
- *  REST trigger fails so the demo still works without a wired loop.
+ *  stream render into that bubble. If the REST trigger fails, the failure is
+ *  surfaced inline on the bubble rather than left streaming forever.
  *
- *  The behavior mode is NOT chosen by the user at send time — the agent / backend
- *  hot-switches it (design/01 §3), and each message carries its own mode, so the
- *  bubble renders per that message's mode. Until the backend advertises a
- *  per-message mode, a new assistant bubble defaults to the current UI mode. */
+ *  The behavior mode is chosen by the agent / backend hot-switch (design/01 §3);
+ *  each message carries the mode in effect when the turn is sent, and the bubble
+ *  renders per that message's mode. We read the current UI mode here and pass it
+ *  on the turn so the backend persists it alongside the message (A3). */
 export function useSendChat(): {
   send: (text: string, image?: string) => Promise<void>;
   sending: boolean;
@@ -46,7 +46,7 @@ export function useSendChat(): {
       beginAssistant(mode, clientId);
 
       try {
-        await chatStream(sessionId, prompt, image);
+        await chatStream(sessionId, prompt, image, mode);
       } catch (err: unknown) {
         // Backend loop not wired or gateway unreachable: show the failure inline
         // rather than leaving a permanently-streaming bubble.