diff --git a/crates/kebab-app/src/lib.rs b/crates/kebab-app/src/lib.rs
index 602fdaf..960442b 100644
--- a/crates/kebab-app/src/lib.rs
+++ b/crates/kebab-app/src/lib.rs
@@ -85,7 +85,7 @@ pub const NO_EXT_SENTINEL: &str = "<no-ext>";
 /// `use kebab_app::AskOpts` keeps working without churn. The struct gained
 /// a `stream_sink` field in P4-3; non-streaming callers (kb-cli today)
 /// pass `stream_sink: None`.
-pub use kebab_rag::AskOpts;
+pub use kebab_rag::{AskOpts, StreamEvent};
 
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub struct DoctorReport {
diff --git a/crates/kebab-tui/src/app.rs b/crates/kebab-tui/src/app.rs
index 44ed73c..1d53d0c 100644
--- a/crates/kebab-tui/src/app.rs
+++ b/crates/kebab-tui/src/app.rs
@@ -186,9 +186,12 @@ impl Default for SearchState {
 /// Ask pane state — owned by p9-3, extended by p9-fb-16 for
 /// multi-turn conversation transcript.
 ///
-/// The worker thread (`thread`) owns the `mpsc::Sender<String>` that
-/// `kebab-app::ask` writes tokens into. The pane keeps the matching
-/// `rx` and drains it once per render frame (no blocking).
+/// The worker thread (`thread`) owns the `mpsc::Sender<kebab_app::StreamEvent>`
+/// that `kebab-app::ask` writes events into. The pane keeps the matching
+/// `rx` and drains it once per render frame (no blocking). Only the
+/// `Token { delta }` variant is consumed for the streaming transcript;
+/// `RetrievalDone` and `Final` are ignored (citations render from
+/// `last_answer` after the worker join).
 ///
 /// p9-fb-16: completed `Turn`s accumulate in `turns`; the worker
 /// passes a snapshot of `turns` as `history` to
@@ -214,7 +217,7 @@ pub struct AskState {
     pub thread: Option<std::thread::JoinHandle<anyhow::Result<kebab_core::Answer>>>,
     /// Token receiver paired with the worker's `Sender`. Drained
     /// every render frame.
-    pub rx: Option<std::sync::mpsc::Receiver<String>>,
+    pub rx: Option<std::sync::mpsc::Receiver<kebab_app::StreamEvent>>,
     /// Vertical scroll offset for the transcript area when content
     /// exceeds the viewport. Only consulted when `follow_tail` is
     /// false; otherwise the renderer overrides this with the
diff --git a/crates/kebab-tui/src/ask.rs b/crates/kebab-tui/src/ask.rs
index dd20917..854a325 100644
--- a/crates/kebab-tui/src/ask.rs
+++ b/crates/kebab-tui/src/ask.rs
@@ -483,7 +483,7 @@ pub fn handle_key_ask(state: &mut App, key: KeyEvent) -> KeyOutcome {
 }
 
 fn spawn_ask_worker(state: &mut App) {
-    let (tx, rx) = mpsc::channel::<String>();
+    let (tx, rx) = mpsc::channel::<kebab_app::StreamEvent>();
     let cfg = state.config.clone();
     let s = state.ask.as_mut().unwrap();
     // p9-fb-10: take() consumes the input in one step (no clone +
@@ -542,8 +542,18 @@ fn make_conversation_id() -> String {
 pub(crate) fn drain_stream(state: &mut App) {
     let Some(s) = state.ask.as_mut() else { return };
     if let Some(rx) = &s.rx {
-        for tok in rx.try_iter() {
-            s.partial.push_str(&tok);
+        for ev in rx.try_iter() {
+            match ev {
+                kebab_app::StreamEvent::Token { delta, .. } => {
+                    s.partial.push_str(&delta);
+                }
+                // p9-fb-33: TUI ignores RetrievalDone (citation
+                // panel renders after completion via `last_answer`)
+                // and Final (the worker thread's join already
+                // delivers the canonical Answer in poll_worker).
+                kebab_app::StreamEvent::RetrievalDone { .. }
+                | kebab_app::StreamEvent::Final { .. } => {}
+            }
         }
     }
 }