From cac662658651ae092cffddfc88f39a775afef942 Mon Sep 17 00:00:00 2001
From: asepharyana <superaseph@gmail.com>
Date: Mon, 20 Jul 2026 14:51:16 +0700
Subject: [PATCH] feat(tui): implement streaming support for LLM responses and
 update transcript handling

---
 apps/interfaces/tui/src/action.rs | 14 ++++++++++++++
 apps/interfaces/tui/src/state.rs  | 11 +++++++++++
 apps/interfaces/tui/src/turn.rs   | 28 ++++++++++++++++++++--------
 3 files changed, 45 insertions(+), 8 deletions(-)

diff --git a/apps/interfaces/tui/src/action.rs b/apps/interfaces/tui/src/action.rs
index 8ad0978..77d6f0a 100644
--- a/apps/interfaces/tui/src/action.rs
+++ b/apps/interfaces/tui/src/action.rs
@@ -158,6 +158,20 @@ pub fn apply_action(state: &mut crate::state::AppStateRest, action: Action) {
                     zesdex_infrastructure::TurnEvent::Error(msg) => {
                         state.toast_error(msg);
                     }
+                    zesdex_infrastructure::TurnEvent::StreamStart => {
+                        state.push_transcript(crate::state::ChatMessageDisplay::new(
+                            zesdex_domain::core::Role::Assistant,
+                            String::new(),
+                        ));
+                    }
+                    zesdex_infrastructure::TurnEvent::StreamToken(text) => {
+                        state.append_to_last_transcript(&text);
+                    }
+                    zesdex_infrastructure::TurnEvent::StreamDone(_msg) => {
+                        // The final message is already accumulated in the transcript.
+                        // We might want to trigger a save or something here, but for UI, we just mark dirty.
+                        state.dirty = true;
+                    }
                     zesdex_infrastructure::TurnEvent::Compacted(msgs) => {
                         if let Some(ref mut rt) = state.session_runtime {
                             rt.messages = msgs;
diff --git a/apps/interfaces/tui/src/state.rs b/apps/interfaces/tui/src/state.rs
index 6e45593..40930d0 100644
--- a/apps/interfaces/tui/src/state.rs
+++ b/apps/interfaces/tui/src/state.rs
@@ -1008,6 +1008,17 @@ impl AppStateRest {
         self.dirty = true;
     }
 
+    /// Append text to the last assistant message in the transcript, if one exists.
+    pub fn append_to_last_transcript(&mut self, text: &str) {
+        if let Some(msg) = self.transcript_cache.messages.back_mut() {
+            if msg.role == zesdex_domain::core::Role::Assistant {
+                msg.content.push_str(text);
+                self.transcript_cache.dirty = true;
+                self.dirty = true;
+            }
+        }
+    }
+
     /// Mark the app state as dirty, triggering a TUI re-render.
     pub fn mark_dirty(&mut self) {
         self.dirty = true;
diff --git a/apps/interfaces/tui/src/turn.rs b/apps/interfaces/tui/src/turn.rs
index 2384299..9c7a213 100644
--- a/apps/interfaces/tui/src/turn.rs
+++ b/apps/interfaces/tui/src/turn.rs
@@ -128,13 +128,27 @@ fn run_turn(
 
         debug!("agent turn iteration {iteration}");
 
-        // Blocking LLM call (reqwest::blocking::Client is sync)
-        let result = client.chat_with_tools_non_streaming(
+        // Stream the LLM response
+        push_event(turn_events, TurnEvent::StreamStart);
+
+        let result = client.chat_with_tools_streaming(
             messages,
             Some(defs.clone()),
-            Some(4096),
             Some(0.7),
-            None,  // no atomic abort flag for the sync API
+            Some(4096),
+            |event| {
+                if abort.load(Ordering::SeqCst) {
+                    return false;
+                }
+                match event {
+                    zesdex_domain::core::StreamEvent::Token(s) | zesdex_domain::core::StreamEvent::Reasoning(s) => {
+                        push_event(turn_events, TurnEvent::StreamToken(s.clone()));
+                    }
+                    _ => {}
+                }
+                true
+            },
+            Some(&abort),
         );
 
         match result {
@@ -142,6 +156,8 @@ fn run_turn(
                 let content = assistant_msg.content.clone().unwrap_or_default();
                 let tool_calls = assistant_msg.tool_calls.clone().unwrap_or_default();
 
+                push_event(turn_events, TurnEvent::StreamDone(assistant_msg.clone()));
+
                 if let Some((tokens_in, tokens_out)) = usage {
                     push_event(turn_events, TurnEvent::Usage {
                         tokens_in,
@@ -149,10 +165,6 @@ fn run_turn(
                     });
                 }
 
-                if !content.is_empty() {
-                    push_event(turn_events, TurnEvent::AssistantMessage(assistant_msg.clone()));
-                }
-
                 if tool_calls.is_empty() {
                     messages.push(ChatMessage::assistant(Some(content)));
                     break;