feat: implement message deduplication and LLM config improvements

- Deduplicate consecutive messages with same role in conversation history
- Add n_predict configuration option for LLM server
- Prevent duplicate message storage in session manager
- Update announcement schedule timing from 37 to 55 minutes
- Add default n_predict value in default bot config

src/bot/mod.rs

@@ -553,10 +553,21 @@ impl BotOrchestrator {
             session_manager.get_conversation_history(session.id, session.user_id)?
         };
 
-        let recent_history = if history.len() > 10 {
+        // Deduplicate consecutive messages from same role
-            &history[history.len() - 10..]
+        let mut deduped_history: Vec<(String, String)> = Vec::new();
+        let mut last_role = None;
+        for (role, content) in history.iter() {
+            if last_role != Some(role) || !deduped_history.is_empty() && 
+               content != &deduped_history.last().unwrap().1 {
+                deduped_history.push((role.clone(), content.clone()));
+                last_role = Some(role);
+            }
+        }
+
+        let recent_history = if deduped_history.len() > 10 {
+            &deduped_history[deduped_history.len() - 10..]
         } else {
-            &history[..]
+            &deduped_history[..]
         };
 
         for (role, content) in recent_history {
@@ -1331,7 +1342,7 @@ async fn websocket_handler(
                     match orchestrator.process_message(user_message.clone()).await {
                         Ok(_) => (),
                         Err(e) => {
-                            error!("Failed to process message: {}", e);
+                        error!("Failed to process message: {}", e);
                             // Fall back to streaming if processing fails
                             if let Err(e) = orchestrator.stream_response(user_message, tx.clone()).await {
                                 error!("Failed to stream response: {}", e);

src/llm/local.rs

@@ -200,6 +200,7 @@ pub async fn start_llm_server(
     let mlock = config_manager.get_config(&default_bot_id, "llm-server-mlock", None).unwrap_or("true".to_string());
     let no_mmap = config_manager.get_config(&default_bot_id, "llm-server-no-mmap", None).unwrap_or("true".to_string());
     let gpu_layers = config_manager.get_config(&default_bot_id, "llm-server-gpu-layers", None).unwrap_or("20".to_string());
+    let n_predict = config_manager.get_config(&default_bot_id, "llm-server-n-predict", None).unwrap_or("50".to_string());
 
     // Build command arguments dynamically
     let mut args = format!(
@@ -222,6 +223,9 @@ pub async fn start_llm_server(
     if no_mmap == "true" {
         args.push_str(" --no-mmap");
     }
+    if n_predict != "0" {
+        args.push_str(&format!(" --n-predict {}", n_predict));
+    }
 
     if cfg!(windows) {
         let mut cmd = tokio::process::Command::new("cmd");

src/session/mod.rs

@@ -189,6 +189,22 @@ impl SessionManager {
     ) -> Result<(), Box<dyn Error + Send + Sync>> {
         use crate::shared::models::message_history::dsl::*;
 
+        // Check if this exact message already exists
+        let exists = message_history
+            .filter(session_id.eq(sess_id))
+            .filter(user_id.eq(uid))
+            .filter(role.eq(ro))
+            .filter(content_encrypted.eq(content))
+            .filter(message_type.eq(msg_type))
+            .select(id)
+            .first::<Uuid>(&mut self.conn)
+            .optional()?;
+
+        if exists.is_some() {
+            debug!("Duplicate message detected, skipping save");
+            return Ok(());
+        }
+
         let next_index = message_history
             .filter(session_id.eq(sess_id))
             .count()

templates/announcements.gbai/announcements.gbdialog/update-summary.bas

@@ -1,4 +1,4 @@
-SET_SCHEDULE "37 * * * *"
+SET_SCHEDULE "55 * * * *"
 
 let text = GET "announcements.gbkb/news/news.pdf"
 let resume = LLM "In a few words, resume this: " + text

templates/default.gbai/default.gbot/config.csv

@@ -18,6 +18,7 @@ llm-server-port,8081
 llm-server-gpu-layers,0
 llm-server-n-moe,0
 llm-server-ctx-size,512
+llm-server-n-predict, 50
 llm-server-parallel,6
 llm-server-cont-batching,true
 llm-server-mlock,false