full chat implemented

src/app.rs

@@ -457,6 +457,7 @@ enum GenerationCommand {
         engine: crate::settings::EngineSettings,
         turn: crate::settings::TurnSettings,
         messages: Vec<ChatTurn>,
+        checkpoint: PathBuf,
         idle_timeout: Duration,
         cancel: Arc<AtomicBool>,
     },
@@ -972,9 +973,23 @@ impl App {
                         Some("Stop the active generation before deleting its project.".into());
                     return Task::none();
                 }
+                let checkpoint_ids = self
+                    .projects
+                    .iter()
+                    .find(|item| item.project.id == project_id)
+                    .map(|item| {
+                        item.sessions
+                            .iter()
+                            .map(|session| session.id)
+                            .collect::<Vec<_>>()
+                    })
+                    .unwrap_or_default();
                 if let Some(database) = &mut self.database {
                     match database.delete_project(project_id) {
                         Ok(()) => {
+                            for session_id in checkpoint_ids {
+                                let _ = fs::remove_file(session_checkpoint_path(session_id));
+                            }
                             if self.selected_project == Some(project_id) {
                                 self.selected_project = None;
                                 self.selected_session = None;
@@ -1036,6 +1051,7 @@ impl App {
                 if let Some(database) = &mut self.database {
                     match database.delete_session(session_id) {
                         Ok(()) => {
+                            let _ = fs::remove_file(session_checkpoint_path(session_id));
                             if self.selected_session == Some(session_id) {
                                 self.selected_session = None;
                                 self.conversation.clear();
@@ -1465,6 +1481,7 @@ impl App {
                 engine: effective.engine,
                 turn: effective.turn,
                 messages,
+                checkpoint: session_checkpoint_path(session_id),
                 idle_timeout,
                 cancel: Arc::clone(&cancel),
             };
@@ -1594,6 +1611,7 @@ fn spawn_generation_worker() -> Result<GenerationWorker, String> {
                         engine,
                         turn,
                         messages,
+                        checkpoint,
                         idle_timeout: requested_timeout,
                         cancel,
                     }) => {
@@ -1614,6 +1632,7 @@ fn spawn_generation_worker() -> Result<GenerationWorker, String> {
                         }
                         if let Some((_, generator)) = &mut loaded {
                             let result = generator.generate(
+                                &checkpoint,
                                 &messages,
                                 &turn,
                                 &cancel,
@@ -1687,6 +1706,12 @@ fn models_path() -> PathBuf {
     application_support_path().join("models")
 }
 
+fn session_checkpoint_path(session_id: i32) -> PathBuf {
+    application_support_path()
+        .join("kv-cache")
+        .join(format!("{session_id}.bin"))
+}
+
 pub(crate) fn app_icon() -> window::Icon {
     let decoder = png::Decoder::new(std::io::Cursor::new(include_bytes!(
         "../assets/app-icon.png"