fix: update Minimax and DeepSeek handlers to strip unclosed thinking tags

src/llm/episodic_memory.rs

@@ -44,9 +44,9 @@ async fn process_episodic_memory(
 
         let history_to_keep = config_manager
             .get_config(&session.bot_id, "history-limit", Some("20")) // Respect history-limit if present
-            .unwrap_or_else(|_| "20".to_string())
+            .unwrap_or_else(|_| "120".to_string())
             .parse::<usize>()
-            .unwrap_or(20);
+            .unwrap_or(120);
 
         if threshold == 0 {
             continue; // Skip this session, episodic memory is disabled for this bot

src/llm/llm_models/deepseek_r3.rs

@@ -6,11 +6,24 @@ static THINK_TAG_REGEX: LazyLock<Result<Regex, regex::Error>> =
     LazyLock::new(|| Regex::new(r"(?s)<think>.*?</think>"));
 
 pub fn strip_think_tags(content: &str) -> String {
-    if let Ok(re) = &*THINK_TAG_REGEX {
+    // We want to strip <think>...</think> OR <think> until end of string (streaming)
-        re.replace_all(content, "").to_string()
+    let mut result = content.to_string();
-    } else {
+    if let Some(start_idx) = result.find("<think>") {
-        content.to_string()
+        if let Some(end_idx) = result[start_idx..].find("</think>") {
+            // Case 1: Fully enclosed
+            result = format!(
+                "{}{}",
+                &result[..start_idx],
+                &result[start_idx + end_idx + 8..]
+            );
+            // Recursive call to catch multiple blocks
+            return strip_think_tags(&result);
+        } else {
+            // Case 2: Unclosed (streaming)
+            result = result[..start_idx].to_string();
+        }
     }
+    result
 }
 
 #[derive(Debug)]

src/llm/llm_models/minimax.rs

@@ -3,7 +3,7 @@ use super::ModelHandler;
 pub fn strip_think_tags(content: &str) -> String {
     let mut result = content.to_string();
 
-    // Chinese: （分析）...（/分析）
+    // Chinese: （分析）...（/分析） or unclosed （分析）...
     while let Some(start_idx) = result.find("（分析）") {
         if let Some(end_idx) = result[start_idx..].find("（/分析）") {
             result = format!(
@@ -12,11 +12,13 @@ pub fn strip_think_tags(content: &str) -> String {
                 &result[start_idx + end_idx + 4..]
             );
         } else {
+            // Unclosed - strip to the end
+            result = result[..start_idx].to_string();
             break;
         }
     }
 
-    // English: <think>...</think>
+    // English: <think>...</think> or unclosed <think>...
     while let Some(start_idx) = result.find("<think>") {
         if let Some(end_idx) = result[start_idx..].find("</think>") {
             result = format!(
@@ -25,11 +27,13 @@ pub fn strip_think_tags(content: &str) -> String {
                 &result[start_idx + end_idx + 8..]
             );
         } else {
+            // Unclosed - strip to the end
+            result = result[..start_idx].to_string();
             break;
         }
     }
 
-    // Chinese alternative: 【分析】...【/分析】
+    // Chinese alternative: 【分析】...【/分析】 or unclosed 【分析】...
     while let Some(start_idx) = result.find("【分析】") {
         if let Some(end_idx) = result[start_idx..].find("【/分析】") {
             result = format!(
@@ -38,6 +42,8 @@ pub fn strip_think_tags(content: &str) -> String {
                 &result[start_idx + end_idx + 5..]
             );
         } else {
+            // Unclosed - strip to the end
+            result = result[..start_idx].to_string();
             break;
         }
     }

src/llm/llm_models/mod.rs

@@ -35,7 +35,7 @@ pub fn get_handler(model_path: &str) -> Box<dyn ModelHandler> {
         Box::new(gpt_oss_120b::GptOss120bHandler::new())
     } else if path.contains("20b") {
         Box::new(gpt_oss_20b::GptOss20bHandler)
-    } else if path.contains("minimax") || path.contains("minimax-m") {
+    } else if path.contains("minimax") || path.contains("minimax-m") || path.contains("kimi") {
         Box::new(minimax::MinimaxHandler::new())
     } else {
         Box::new(PassthroughHandler)