diff --git a/profiler/src/profiler/TracyConfig.cpp b/profiler/src/profiler/TracyConfig.cpp
index 47b15c59..cb076614 100644
--- a/profiler/src/profiler/TracyConfig.cpp
+++ b/profiler/src/profiler/TracyConfig.cpp
@@ -52,6 +52,8 @@ void LoadConfig()
     if( ini_sget( ini, "llm", "limitToolReplySize", "%d", &v ) ) s_config.llmLimitToolReplySize = v;
     if( ini_sget( ini, "llm", "maxToolReplySizeValue", "%d", &v ) ) s_config.llmMaxToolReplySizeValue = v;
     if( ini_sget( ini, "llm", "separateFastModel", "%d", &v ) ) s_config.llmSeparateFastModel = v;
+    if( ini_sget( ini, "llm", "summary", "%d", &v ) ) s_config.llmSummary = v;
+    if( ini_sget( ini, "llm", "suggestion", "%d", &v ) ) s_config.llmSuggestion = v;
 
     ini_free( ini );
 }
@@ -105,6 +107,8 @@ bool SaveConfig()
     fprintf( f, "limitToolReplySize = %i\n", (int)s_config.llmLimitToolReplySize );
     fprintf( f, "maxToolReplySizeValue = %i\n", s_config.llmMaxToolReplySizeValue );
     fprintf( f, "separateFastModel = %i\n", (int)s_config.llmSeparateFastModel );
+    fprintf( f, "summary = %i\n", (int)s_config.llmSummary );
+    fprintf( f, "suggestion = %i\n", (int)s_config.llmSuggestion );
 
     fclose( f );
     return true;
diff --git a/profiler/src/profiler/TracyConfig.hpp b/profiler/src/profiler/TracyConfig.hpp
index 461ae45b..ab76d22a 100644
--- a/profiler/src/profiler/TracyConfig.hpp
+++ b/profiler/src/profiler/TracyConfig.hpp
@@ -48,6 +48,8 @@ struct Config
     bool llmAnnotateCallstacks = false;
     bool llmLimitToolReplySize = false;
     int llmMaxToolReplySizeValue = 48*1024;
+    bool llmSummary = true;
+    bool llmSuggestion = true;
 };
 
 extern Config s_config;
diff --git a/profiler/src/profiler/TracyLlm.cpp b/profiler/src/profiler/TracyLlm.cpp
index b558ed34..70e18dd4 100644
--- a/profiler/src/profiler/TracyLlm.cpp
+++ b/profiler/src/profiler/TracyLlm.cpp
@@ -325,6 +325,18 @@ void TracyLlm::Draw()
             SaveConfig();
         }
 
+        if( ImGui::Checkbox( ICON_FA_HAND_POINT_RIGHT " Show summary", &s_config.llmSummary ) )
+        {
+            SaveConfig();
+        }
+        ImGui::SameLine();
+        ImGui::Spacing();
+        ImGui::SameLine();
+        if( ImGui::Checkbox( ICON_FA_COMMENT_DOTS " Chat suggestion", &s_config.llmSuggestion ) )
+        {
+            SaveConfig();
+        }
+
         if( ImGui::TreeNode( "Advanced" ) )
         {
             if( responding ) ImGui::BeginDisabled();
@@ -1077,7 +1089,7 @@ void TracyLlm::SendMessage()
     std::unique_lock lock( m_chatLock );
     ManageContext();
     auto chat = m_chat;
-    bool needSummary = m_summary.empty();
+    bool needSummary = m_summary.empty() && s_config.llmSummary;
     lock.unlock();
 
     for( auto& msg : chat )
@@ -1330,30 +1342,33 @@ bool TracyLlm::OnResponse( const nlohmann::json& json )
                 if( msg.contains( "model" ) ) msg.erase( "model" );
             }
 
-            auto suggestionQuery = chat;
-            suggestionQuery.push_back( nlohmann::json {
-                {"role", "user"},
-                {"content", "Based on this conversation, suggest one useful follow-up question the user might want to ask next. It should be relevant, actionable, and something the user would genuinely want to explore. Reply with ONLY the question text, in the user's language, under 80 characters."}
-            } );
-            const int chatId = m_chatId.load( std::memory_order_acquire );
-            QueueFastMessageLocking( suggestionQuery, [this, chatId]( const nlohmann::json& res ) {
-                if( m_chatId.load( std::memory_order_acquire ) != chatId ) return;
-                if( res.contains( "choices" ) )
-                {
-                    auto& choices = res["choices"];
-                    if( choices.is_array() && !choices.empty() )
+            if( s_config.llmSuggestion )
+            {
+                auto suggestionQuery = chat;
+                suggestionQuery.push_back( nlohmann::json {
+                    {"role", "user"},
+                    {"content", "Based on this conversation, suggest one useful follow-up question the user might want to ask next. It should be relevant, actionable, and something the user would genuinely want to explore. Reply with ONLY the question text, in the user's language, under 80 characters."}
+                } );
+                const int chatId = m_chatId.load( std::memory_order_acquire );
+                QueueFastMessageLocking( suggestionQuery, [this, chatId]( const nlohmann::json& res ) {
+                    if( m_chatId.load( std::memory_order_acquire ) != chatId ) return;
+                    if( res.contains( "choices" ) )
                     {
-                        if( choices[0].contains( "message" ) && choices[0]["message"].contains( "content" ) )
+                        auto& choices = res["choices"];
+                        if( choices.is_array() && !choices.empty() )
                         {
-                            auto str = choices[0]["message"]["content"].get<std::string>();
-                            std::ranges::replace( str, '\n', ' ' );
-                            std::ranges::replace( str, '\r', ' ' );
-                            std::lock_guard lock( m_chatLock );
-                            m_suggestion = std::move( str );
+                            if( choices[0].contains( "message" ) && choices[0]["message"].contains( "content" ) )
+                            {
+                                auto str = choices[0]["message"]["content"].get<std::string>();
+                                std::ranges::replace( str, '\n', ' ' );
+                                std::ranges::replace( str, '\r', ' ' );
+                                std::lock_guard lock( m_chatLock );
+                                m_suggestion = std::move( str );
+                            }
                         }
                     }
-                }
-            } );
+                } );
+            }
 
             jobsLock.lock();
             m_focusInput = true;