Detect reasoning withing chat in ollama chat completion.

Fix minor bug in Http async.

src/eepp/network/http.cpp

@@ -1179,13 +1179,6 @@ void Http::AsyncRequest::run() {
 		eeSAFE_DELETE( mStream );
 	}
 
-	// The Async Request destroys the socket used to create the request
-	if ( mHttp->mConnection ) {
-		HttpConnection* connection = mHttp->mConnection;
-		eeSAFE_DELETE( connection );
-		mHttp->mConnection = NULL;
-	}
-
 	mRunning = false;
 }
 

src/tools/ecode/plugins/aiassistant/chatui.cpp

@@ -931,13 +931,13 @@ LLMModel LLMChatUI::findModel( const std::string& provider, const std::string& m
 			return *modelIt;
 		}
 	}
-	if ( provider == "openai" && model == "gpt-4o" )
+	if ( provider == "openai" && model == "gpt-5" )
 		return mCurModel; // Do not stack-overflow if something is really wrong
 	return getDefaultModel();
 }
 
 LLMModel LLMChatUI::getDefaultModel() {
-	return findModel( "openai", "gpt-4o" );
+	return findModel( "openai", "gpt-5" );
 }
 
 std::string LLMChatUI::getNewFilePath( const std::string& uuid, const std::string& summary,
@@ -970,6 +970,12 @@ std::string LLMChatUI::prepareApiUrl( const std::string& apiKey ) {
 	return url;
 }
 
+static bool allNewLines( const std::string& s ) {
+	if ( s.empty() )
+		return false;
+	return std::all_of( s.begin(), s.end(), []( char c ) { return c == '\n'; } );
+}
+
 void LLMChatUI::doRequest() {
 	if ( mRequest )
 		return;
@@ -1003,7 +1009,8 @@ void LLMChatUI::doRequest() {
 
 	mRequest->streamedResponseCb = [this, editor, thinking, thinkingID]( const std::string& chunk,
 																		 bool fromReasoning ) {
-		if ( fromReasoning )
+		if ( fromReasoning ||
+			 ( mRequest && mRequest->getResponse().empty() && allNewLines( chunk ) ) )
 			return;
 		auto conversation = chunk;
 		editor->runOnMainThread(
@@ -1086,7 +1093,7 @@ void LLMChatUI::doRequest() {
 						String::trimInPlace( mSummary, ' ' );
 						String::trimInPlace( mSummary, '"' );
 					} else {
-						// TODO: Implement generating a summary base on the user prompt (take the
+						// TODO: Implement generating a summary based on the user prompt (take the
 						// first few words)
 						mSummary = i18n( "untitled_conversation", "Untitled Conversation" );
 					}

src/tools/ecode/plugins/aiassistant/llmchatcompletionrequest.cpp

@@ -48,9 +48,21 @@ LLMChatCompletionRequest::LLMChatCompletionRequest( const std::string& uri, cons
 						std::string delta = msg["content"];
 						if ( delta.empty() )
 							return;
+						if ( mFirstMessage ) {
+							if ( String::startsWith( delta, "<think>" ) )
+								mReasoning = true;
+							mFirstMessage = false;
+						}
+
 						if ( streamedResponseCb )
-							streamedResponseCb( delta, false );
-						mResponse += std::move( delta );
+							streamedResponseCb( delta, mReasoning );
+						if ( mReasoning ) {
+							if ( String::endsWith( delta, "</think>" ) )
+								mReasoning = false;
+							mReasoningResponse += std::move( delta );
+						} else {
+							mResponse += std::move( delta );
+						}
 					}
 				}
 
@@ -70,6 +82,7 @@ LLMChatCompletionRequest::LLMChatCompletionRequest( const std::string& uri, cons
 						std::string delta = choice["delta"]["reasoning_content"];
 						if ( streamedResponseCb )
 							streamedResponseCb( delta, true );
+						mReasoningResponse += std::move( delta );
 					} else if ( choice["delta"].contains( "content" ) ) {
 						std::string delta = choice["delta"]["content"];
 						if ( streamedResponseCb )

src/tools/ecode/plugins/aiassistant/llmchatcompletionrequest.hpp

@@ -40,14 +40,19 @@ class LLMChatCompletionRequest {
 
 	const std::string& getResponse() const { return mResponse; }
 
+	const std::string& getReasoningResponse() const { return mReasoningResponse; }
+
   protected:
 	URI mUrl;
 	Http mHttp;
 	Http::Request mRequest;
 	IOStreamString mStream;
+	std::string mReasoningResponse;
 	std::string mResponse;
 	size_t mReadBytes{ 0 };
 	bool mCancel{ false };
+	bool mFirstMessage{ true };
+	bool mReasoning{ false };
 };
 
 } // namespace ecode