feat: Add using vector and chunks in Context Manager

CMakeLists.txt

@@ -8,7 +8,6 @@ set(CMAKE_AUTOUIC ON)
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
-set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 find_package(QtCreator REQUIRED COMPONENTS Core)
 find_package(Qt6 COMPONENTS Core Gui Quick Widgets Network REQUIRED)
@@ -44,30 +43,26 @@ add_qtc_plugin(QodeAssist
     LLMClientInterface.hpp LLMClientInterface.cpp
     templates/Templates.hpp
     templates/CodeLlamaFim.hpp
-    templates/Ollama.hpp
-    templates/Claude.hpp
-    templates/OpenAI.hpp
-    templates/MistralAI.hpp
     templates/StarCoder2Fim.hpp
-    # templates/DeepSeekCoderFim.hpp
-    # templates/CustomFimTemplate.hpp
+    templates/DeepSeekCoderFim.hpp
+    templates/CustomFimTemplate.hpp
     templates/Qwen.hpp
-    templates/OpenAICompatible.hpp
+    templates/Ollama.hpp
+    templates/BasicChat.hpp
     templates/Llama3.hpp
     templates/ChatML.hpp
     templates/Alpaca.hpp
     templates/Llama2.hpp
+    templates/Claude.hpp
+    templates/OpenAI.hpp
     templates/CodeLlamaQMLFim.hpp
-    templates/GoogleAI.hpp
     providers/Providers.hpp
     providers/OllamaProvider.hpp providers/OllamaProvider.cpp
-    providers/ClaudeProvider.hpp providers/ClaudeProvider.cpp
-    providers/OpenAIProvider.hpp providers/OpenAIProvider.cpp
-    providers/MistralAIProvider.hpp providers/MistralAIProvider.cpp
     providers/LMStudioProvider.hpp providers/LMStudioProvider.cpp
     providers/OpenAICompatProvider.hpp providers/OpenAICompatProvider.cpp
     providers/OpenRouterAIProvider.hpp providers/OpenRouterAIProvider.cpp
-    providers/GoogleAIProvider.hpp providers/GoogleAIProvider.cpp
+    providers/ClaudeProvider.hpp providers/ClaudeProvider.cpp
+    providers/OpenAIProvider.hpp providers/OpenAIProvider.cpp
     QodeAssist.qrc
     LSPCompletion.hpp
     LLMSuggestion.hpp LLMSuggestion.cpp

ChatView/ChatRootView.cpp

@@ -24,21 +24,24 @@
 #include <QFileDialog>
 #include <QMessageBox>
 
+#include <coreplugin/editormanager/editormanager.h>
 #include <coreplugin/icore.h>
 #include <projectexplorer/project.h>
 #include <projectexplorer/projectexplorer.h>
 #include <projectexplorer/projectmanager.h>
+#include <projectexplorer/projecttree.h>
 #include <utils/theme/theme.h>
 #include <utils/utilsicons.h>
-#include <coreplugin/editormanager/editormanager.h>
 
 #include "ChatAssistantSettings.hpp"
 #include "ChatSerializer.hpp"
 #include "GeneralSettings.hpp"
 #include "Logger.hpp"
 #include "ProjectSettings.hpp"
-#include "context/TokenUtils.hpp"
 #include "context/ContextManager.hpp"
+#include "context/FileChunker.hpp"
+#include "context/RAGManager.hpp"
+#include "context/TokenUtils.hpp"
 
 namespace QodeAssist::Chat {
 
@@ -447,6 +450,69 @@ void ChatRootView::openChatHistoryFolder()
     QDesktopServices::openUrl(url);
 }
 
+// ChatRootView.cpp
+
+void ChatRootView::testRAG(const QString &message)
+{
+    auto project = ProjectExplorer::ProjectTree::currentProject();
+    if (!project) {
+        qDebug() << "No active project found";
+        return;
+    }
+
+    const QString TEST_QUERY = message;
+
+    qDebug() << "Starting RAG test with query:";
+    qDebug() << TEST_QUERY;
+    qDebug() << "\nFirst, processing project files...";
+
+    auto files = Context::ContextManager::instance().getProjectSourceFiles(project);
+    // Было: auto future = Context::RAGManager::instance().processFiles(project, files);
+    // Стало:
+    auto future = Context::RAGManager::instance().processProjectFiles(project, files);
+
+    connect(
+        &Context::RAGManager::instance(),
+        &Context::RAGManager::vectorizationProgress,
+        this,
+        [](int processed, int total) {
+            qDebug() << QString("Vectorization progress: %1 of %2 files").arg(processed).arg(total);
+        });
+
+    connect(
+        &Context::RAGManager::instance(),
+        &Context::RAGManager::vectorizationFinished,
+        this,
+        [this, project, TEST_QUERY]() {
+            qDebug() << "\nVectorization completed. Starting similarity search...\n";
+            // Было: Context::RAGManager::instance().searchSimilarDocuments(TEST_QUERY, project, 5);
+            // Стало:
+            auto future = Context::RAGManager::instance().findRelevantChunks(TEST_QUERY, project, 5);
+            future.then([](const QList<Context::RAGManager::ChunkSearchResult> &results) {
+                qDebug() << "Found" << results.size() << "relevant chunks:";
+                for (const auto &result : results) {
+                    qDebug() << "File:" << result.filePath;
+                    qDebug() << "Lines:" << result.startLine << "-" << result.endLine;
+                    qDebug() << "Score:" << result.combinedScore;
+                    qDebug() << "Content:" << result.content;
+                    qDebug() << "---";
+                }
+            });
+        });
+}
+
+void ChatRootView::testChunking()
+{
+    auto project = ProjectExplorer::ProjectTree::currentProject();
+    if (!project) {
+        qDebug() << "No active project found";
+        return;
+    }
+
+    Context::FileChunker::ChunkingConfig config;
+    Context::ContextManager::instance().testProjectChunks(project, config);
+}
+
 void ChatRootView::updateInputTokensCount()
 {
     int inputTokens = m_messageTokensCount;

ChatView/ChatRootView.hpp

@@ -65,6 +65,8 @@ public:
     Q_INVOKABLE void calculateMessageTokensCount(const QString &message);
     Q_INVOKABLE void setIsSyncOpenFiles(bool state);
     Q_INVOKABLE void openChatHistoryFolder();
+    Q_INVOKABLE void testRAG(const QString &message);
+    Q_INVOKABLE void testChunking();
 
     Q_INVOKABLE void updateInputTokensCount();
     int inputTokensCount() const;

ChatView/ClientInterface.cpp

@@ -93,47 +93,51 @@ void ClientInterface::sendMessage(
     }
 
     LLMCore::ContextData context;
+    context.prefix = message;
+    context.suffix = "";
 
-    if (chatAssistantSettings.useSystemPrompt()) {
-        QString systemPrompt = chatAssistantSettings.systemPrompt();
-        if (!linkedFiles.isEmpty()) {
-            systemPrompt = getSystemPromptWithLinkedFiles(systemPrompt, linkedFiles);
-        }
-        context.systemPrompt = systemPrompt;
+    QString systemPrompt;
+    if (chatAssistantSettings.useSystemPrompt())
+        systemPrompt = chatAssistantSettings.systemPrompt();
+
+    if (!linkedFiles.isEmpty()) {
+        systemPrompt = getSystemPromptWithLinkedFiles(systemPrompt, linkedFiles);
     }
 
-    QVector<LLMCore::Message> messages;
-    for (const auto &msg : m_chatModel->getChatHistory()) {
-        messages.append({msg.role == ChatModel::ChatRole::User ? "user" : "assistant", msg.content});
-    }
-    context.history = messages;
+    QJsonObject providerRequest;
+    providerRequest["model"] = Settings::generalSettings().caModel();
+    providerRequest["stream"] = chatAssistantSettings.stream();
+    providerRequest["messages"] = m_chatModel->prepareMessagesForRequest(systemPrompt);
+
+    if (promptTemplate)
+        promptTemplate->prepareRequest(providerRequest, context);
+    else
+        qWarning("No prompt template found");
+
+    if (provider)
+        provider->prepareRequest(providerRequest, LLMCore::RequestType::Chat);
+    else
+        qWarning("No provider found");
 
     LLMCore::LLMConfig config;
     config.requestType = LLMCore::RequestType::Chat;
     config.provider = provider;
     config.promptTemplate = promptTemplate;
-    if (provider->providerID() == LLMCore::ProviderID::GoogleAI) {
-        QString stream = chatAssistantSettings.stream() ? QString{"streamGenerateContent?alt=sse"}
-                                                        : QString{"generateContent?"};
-        config.url = QUrl(QString("%1/models/%2:%3")
-                              .arg(
-                                  Settings::generalSettings().caUrl(),
-                                  Settings::generalSettings().caModel(),
-                                  stream));
-    } else {
-        config.url
-            = QString("%1%2").arg(Settings::generalSettings().caUrl(), provider->chatEndpoint());
-        config.providerRequest
-            = {{"model", Settings::generalSettings().caModel()},
-               {"stream", chatAssistantSettings.stream()}};
-    }
-
+    config.url = QString("%1%2").arg(Settings::generalSettings().caUrl(), provider->chatEndpoint());
+    config.providerRequest = providerRequest;
+    config.multiLineCompletion = false;
     config.apiKey = provider->apiKey();
 
-    config.provider
-        ->prepareRequest(config.providerRequest, promptTemplate, context, LLMCore::RequestType::Chat);
+    QJsonObject request;
+    request["id"] = QUuid::createUuid().toString();
+
+    auto errors = config.provider->validateRequest(config.providerRequest, promptTemplate->type());
+    if (!errors.isEmpty()) {
+        LOG_MESSAGE("Validate errors for chat request:");
+        LOG_MESSAGES(errors);
+        return;
+    }
 
-    QJsonObject request{{"id", QUuid::createUuid().toString()}};
     m_requestHandler->sendLLMRequest(config, request);
 }
 

ChatView/qml/ChatItem.qml

@@ -27,18 +27,14 @@ Rectangle {
 
     property alias msgModel: msgCreator.model
     property alias messageAttachments: attachmentsModel.model
-    property bool isUserMessage: false
 
     height: msgColumn.implicitHeight + 10
     radius: 8
-    color: isUserMessage ? palette.alternateBase
-                         : palette.base
 
     ColumnLayout {
         id: msgColumn
 
-        x: 5
-        width: parent.width - x
+        width: parent.width
         anchors.verticalCenter: parent.verticalCenter
         spacing: 5
 
@@ -117,17 +113,6 @@ Rectangle {
         }
     }
 
-    Rectangle {
-        id: userMessageMarker
-
-        anchors.verticalCenter: parent.verticalCenter
-        width: 3
-        height: root.height - root.radius
-        color: "#92BD6C"
-        radius: root.radius
-        visible: root.isUserMessage
-    }
-
     component TextComponent : TextBlock {
         required property var itemData
         height: implicitHeight + 10

ChatView/qml/RootItem.qml

@@ -96,7 +96,8 @@ ChatRootView {
                 width: ListView.view.width - scroll.width
                 msgModel: root.chatModel.processMessageContent(model.content)
                 messageAttachments: model.attachments
-                isUserMessage: model.roleType === ChatModel.User
+                color: model.roleType === ChatModel.User ? palette.alternateBase
+                                                         : palette.base
             }
 
             header: Item {
@@ -197,6 +198,8 @@ ChatRootView {
             }
             attachFiles.onClicked: root.showAttachFilesDialog()
             linkFiles.onClicked: root.showLinkFilesDialog()
+            testRag.onClicked: root.testRAG(messageInput.text)
+            testChunks.onClicked: root.testChunking()
         }
     }
 

ChatView/qml/parts/BottomBar.qml

@@ -30,6 +30,8 @@ Rectangle {
     property alias syncOpenFiles: syncOpenFilesId
     property alias attachFiles: attachFilesId
     property alias linkFiles: linkFilesId
+    property alias testRag: testRagId
+    property alias testChunks: testChunksId
 
     color: palette.window.hslLightness > 0.5 ?
                Qt.darker(palette.window, 1.1) :
@@ -91,6 +93,18 @@ Rectangle {
             ToolTip.text: qsTr("Automatically synchronize currently opened files with the model context")
         }
 
+        QoAButton {
+            id: testRagId
+
+            text: qsTr("Test RAG")
+        }
+
+        QoAButton {
+            id: testChunksId
+
+            text: qsTr("Test Chunks")
+        }
+
         Item {
             Layout.fillWidth: true
         }

ConfigurationManager.cpp

@@ -35,28 +35,6 @@ ConfigurationManager &ConfigurationManager::instance()
 void ConfigurationManager::init()
 {
     setupConnections();
-    updateAllTemplateDescriptions();
-}
-
-void ConfigurationManager::updateTemplateDescription(const Utils::StringAspect &templateAspect)
-{
-    LLMCore::PromptTemplate *templ = m_templateManger.getFimTemplateByName(templateAspect.value());
-
-    if (!templ) {
-        return;
-    }
-
-    if (&templateAspect == &m_generalSettings.ccTemplate) {
-        m_generalSettings.ccTemplateDescription.setValue(templ->description());
-    } else if (&templateAspect == &m_generalSettings.caTemplate) {
-        m_generalSettings.caTemplateDescription.setValue(templ->description());
-    }
-}
-
-void ConfigurationManager::updateAllTemplateDescriptions()
-{
-    updateTemplateDescription(m_generalSettings.ccTemplate);
-    updateTemplateDescription(m_generalSettings.caTemplate);
 }
 
 ConfigurationManager::ConfigurationManager(QObject *parent)
@@ -86,14 +64,6 @@ void ConfigurationManager::setupConnections()
     connect(&m_generalSettings.ccPreset1SelectModel, &Button::clicked, this, &Config::selectModel);
     connect(
         &m_generalSettings.ccPreset1SelectTemplate, &Button::clicked, this, &Config::selectTemplate);
-
-    connect(&m_generalSettings.ccTemplate, &Utils::StringAspect::changed, this, [this]() {
-        updateTemplateDescription(m_generalSettings.ccTemplate);
-    });
-
-    connect(&m_generalSettings.caTemplate, &Utils::StringAspect::changed, this, [this]() {
-        updateTemplateDescription(m_generalSettings.caTemplate);
-    });
 }
 
 void ConfigurationManager::selectProvider()
@@ -167,14 +137,9 @@ void ConfigurationManager::selectTemplate()
 
     const bool isCodeCompletion = (settingsButton == &m_generalSettings.ccSelectTemplate);
     const bool isPreset1 = (settingsButton == &m_generalSettings.ccPreset1SelectTemplate);
-    const QString providerName = isCodeCompletion ? m_generalSettings.ccProvider.volatileValue()
-                                 : isPreset1 ? m_generalSettings.ccPreset1Provider.volatileValue()
-                                             : m_generalSettings.caProvider.volatileValue();
-    auto providerID = m_providersManager.getProviderByName(providerName)->providerID();
 
-    const auto templateList = isCodeCompletion || isPreset1
-                                  ? m_templateManger.getFimTemplatesForProvider(providerID)
-                                  : m_templateManger.getChatTemplatesForProvider(providerID);
+    const auto templateList = isCodeCompletion || isPreset1 ? m_templateManger.fimTemplatesNames()
+                                                            : m_templateManger.chatTemplatesNames();
 
     auto &targetSettings = isCodeCompletion ? m_generalSettings.ccTemplate
                            : isPreset1      ? m_generalSettings.ccPreset1Template

ConfigurationManager.hpp

@@ -36,9 +36,6 @@ public:
 
     void init();
 
-    void updateTemplateDescription(const Utils::StringAspect &templateAspect);
-    void updateAllTemplateDescriptions();
-
 public slots:
     void selectProvider();
     void selectModel();

LLMClientInterface.cpp

@@ -27,8 +27,8 @@
 #include <texteditor/textdocument.h>
 
 #include "CodeHandler.hpp"
-#include "context/ContextManager.hpp"
 #include "context/DocumentContextReader.hpp"
+#include "llmcore/MessageBuilder.hpp"
 #include "llmcore/PromptTemplateManager.hpp"
 #include "llmcore/ProvidersManager.hpp"
 #include "logger/Logger.hpp"
@@ -146,13 +146,24 @@ void LLMClientInterface::handleExit(const QJsonObject &request)
     emit finished();
 }
 
+bool QodeAssist::LLMClientInterface::isSpecifyCompletion(const QJsonObject &request)
+{
+    auto &generalSettings = Settings::generalSettings();
+
+    Context::ProgrammingLanguage documentLanguage = getDocumentLanguage(request);
+    Context::ProgrammingLanguage preset1Language = Context::ProgrammingLanguageUtils::fromString(
+        generalSettings.preset1Language.displayForIndex(generalSettings.preset1Language()));
+
+    return generalSettings.specifyPreset1() && documentLanguage == preset1Language;
+}
+
 void LLMClientInterface::handleCompletion(const QJsonObject &request)
 {
-    auto updatedContext = prepareContext(request);
+    const auto updatedContext = prepareContext(request);
     auto &completeSettings = Settings::codeCompletionSettings();
     auto &generalSettings = Settings::generalSettings();
 
-    bool isPreset1Active = Context::ContextManager::instance().isSpecifyCompletion(request);
+    bool isPreset1Active = isSpecifyCompletion(request);
 
     const auto providerName = !isPreset1Active ? generalSettings.ccProvider()
                                                : generalSettings.ccPreset1Provider();
@@ -183,19 +194,14 @@ void LLMClientInterface::handleCompletion(const QJsonObject &request)
     config.requestType = LLMCore::RequestType::CodeCompletion;
     config.provider = provider;
     config.promptTemplate = promptTemplate;
-    // TODO refactor networking
-    if (provider->providerID() == LLMCore::ProviderID::GoogleAI) {
-        QString stream = completeSettings.stream() ? QString{"streamGenerateContent?alt=sse"}
-                                                   : QString{"generateContent?"};
-        config.url = QUrl(QString("%1/models/%2:%3").arg(url, modelName, stream));
-    } else {
-        config.url = QUrl(QString("%1%2").arg(
-            url,
-            promptTemplate->type() == LLMCore::TemplateType::FIM ? provider->completionEndpoint()
-                                                                 : provider->chatEndpoint()));
-        config.providerRequest = {{"model", modelName}, {"stream", completeSettings.stream()}};
-    }
+    config.url = QUrl(QString("%1%2").arg(
+        url,
+        promptTemplate->type() == LLMCore::TemplateType::Fim ? provider->completionEndpoint()
+                                                             : provider->chatEndpoint()));
     config.apiKey = provider->apiKey();
+
+    config.providerRequest = {{"model", modelName}, {"stream", completeSettings.stream()}};
+
     config.multiLineCompletion = completeSettings.multiLineCompletion();
 
     const auto stopWords = QJsonArray::fromStringList(config.promptTemplate->stopWords());
@@ -204,35 +210,28 @@ void LLMClientInterface::handleCompletion(const QJsonObject &request)
 
     QString systemPrompt;
     if (completeSettings.useSystemPrompt())
-        systemPrompt.append(completeSettings.useUserMessageTemplateForCC()
-                                    && promptTemplate->type() == LLMCore::TemplateType::Chat
-                                ? completeSettings.systemPromptForNonFimModels()
-                                : completeSettings.systemPrompt());
-    if (updatedContext.fileContext.has_value())
-        systemPrompt.append(updatedContext.fileContext.value());
+        systemPrompt.append(completeSettings.systemPrompt());
+    if (!updatedContext.fileContext.isEmpty())
+        systemPrompt.append(updatedContext.fileContext);
 
-    updatedContext.systemPrompt = systemPrompt;
-
-    if (promptTemplate->type() == LLMCore::TemplateType::Chat) {
-        QString userMessage;
-        if (completeSettings.useUserMessageTemplateForCC()) {
-            userMessage = completeSettings.processMessageToFIM(
-                updatedContext.prefix.value_or(""), updatedContext.suffix.value_or(""));
-        } else {
-            userMessage = updatedContext.prefix.value_or("") + updatedContext.suffix.value_or("");
-        }
-
-        // TODO refactor add message
-        QVector<LLMCore::Message> messages;
-        messages.append({"user", userMessage});
-        updatedContext.history = messages;
+    QString userMessage;
+    if (completeSettings.useUserMessageTemplateForCC() && promptTemplate->type() == LLMCore::TemplateType::Chat) {
+        userMessage = completeSettings.userMessageTemplateForCC().arg(updatedContext.prefix, updatedContext.suffix);
+    } else {
+        userMessage = updatedContext.prefix;
     }
 
-    config.provider->prepareRequest(
+    auto message = LLMCore::MessageBuilder()
+                       .addSystemMessage(systemPrompt)
+                       .addUserMessage(userMessage)
+                       .addSuffix(updatedContext.suffix)
+                       .addTokenizer(promptTemplate);
+
+    message.saveTo(
         config.providerRequest,
-        promptTemplate,
-        updatedContext,
-        LLMCore::RequestType::CodeCompletion);
+        providerName == "Ollama" ? LLMCore::ProvidersApi::Ollama : LLMCore::ProvidersApi::OpenAI);
+
+    config.provider->prepareRequest(config.providerRequest, LLMCore::RequestType::CodeCompletion);
 
     auto errors = config.provider->validateRequest(config.providerRequest, promptTemplate->type());
     if (!errors.isEmpty()) {
@@ -267,11 +266,29 @@ LLMCore::ContextData LLMClientInterface::prepareContext(const QJsonObject &reque
     return reader.prepareContext(lineNumber, cursorPosition);
 }
 
+Context::ProgrammingLanguage LLMClientInterface::getDocumentLanguage(const QJsonObject &request) const
+{
+    QJsonObject params = request["params"].toObject();
+    QJsonObject doc = params["doc"].toObject();
+    QString uri = doc["uri"].toString();
+
+    Utils::FilePath filePath = Utils::FilePath::fromString(QUrl(uri).toLocalFile());
+    TextEditor::TextDocument *textDocument = TextEditor::TextDocument::textDocumentForFilePath(
+        filePath);
+
+    if (!textDocument) {
+        LOG_MESSAGE("Error: Document is not available for" + filePath.toString());
+        return Context::ProgrammingLanguage::Unknown;
+    }
+
+    return Context::ProgrammingLanguageUtils::fromMimeType(textDocument->mimeType());
+}
+
 void LLMClientInterface::sendCompletionToClient(const QString &completion,
                                                 const QJsonObject &request,
                                                 bool isComplete)
 {
-    bool isPreset1Active = Context::ContextManager::instance().isSpecifyCompletion(request);
+    bool isPreset1Active = isSpecifyCompletion(request);
 
     auto templateName = !isPreset1Active ? Settings::generalSettings().ccTemplate()
                                          : Settings::generalSettings().ccPreset1Template();
@@ -289,18 +306,18 @@ void LLMClientInterface::sendCompletionToClient(const QString &completion,
     QJsonArray completions;
     QJsonObject completionItem;
 
-    LOG_MESSAGE(QString("Completions before filter: \n%1").arg(completion));
-
     QString processedCompletion
         = promptTemplate->type() == LLMCore::TemplateType::Chat
                   && Settings::codeCompletionSettings().smartProcessInstuctText()
-                                      ? CodeHandler::processText(completion)
-                                      : completion;
+              ? CodeHandler::processText(completion)
+              : completion;
 
     completionItem[LanguageServerProtocol::textKey] = processedCompletion;
     QJsonObject range;
     range["start"] = position;
-    range["end"] = position;
+    QJsonObject end = position;
+    end["character"] = position["character"].toInt() + processedCompletion.length();
+    range["end"] = end;
     completionItem[LanguageServerProtocol::rangeKey] = range;
     completionItem[LanguageServerProtocol::positionKey] = position;
     completions.append(completionItem);
@@ -313,7 +330,7 @@ void LLMClientInterface::sendCompletionToClient(const QString &completion,
             .arg(QString::fromUtf8(QJsonDocument(completions).toJson(QJsonDocument::Indented))));
 
     LOG_MESSAGE(QString("Full response: \n%1")
-            .arg(QString::fromUtf8(QJsonDocument(response).toJson(QJsonDocument::Indented))));
+                   .arg(QString::fromUtf8(QJsonDocument(response).toJson(QJsonDocument::Indented))));
 
     QString requestId = request["id"].toString();
     endTimeMeasurement(requestId);

LLMClientInterface.hpp

@@ -61,6 +61,8 @@ private:
 
     LLMCore::ContextData prepareContext(
         const QJsonObject &request, const QStringView &accumulatedCompletion = QString{});
+    Context::ProgrammingLanguage getDocumentLanguage(const QJsonObject &request) const;
+    bool isSpecifyCompletion(const QJsonObject &request);
 
     LLMCore::RequestHandler m_requestHandler;
     QElapsedTimer m_completionTimer;

QodeAssist.json.in

@@ -1,7 +1,7 @@
 {
     "Id" : "qodeassist",
     "Name" : "QodeAssist",
-    "Version" : "0.5.0",
+    "Version" : "0.4.13",
     "Vendor" : "Petr Mironychev",
     "VendorId" : "petrmironychev",
     "Copyright" : "(C) ${IDE_COPYRIGHT_YEAR} Petr Mironychev, (C) ${IDE_COPYRIGHT_YEAR} The Qt Company Ltd",

context/CMakeLists.txt

@@ -5,11 +5,20 @@ add_library(Context STATIC
     ContentFile.hpp
     TokenUtils.hpp TokenUtils.cpp
     ProgrammingLanguage.hpp ProgrammingLanguage.cpp
+    RAGManager.hpp RAGManager.cpp
+    RAGStorage.hpp RAGStorage.cpp
+    RAGData.hpp
+    RAGVectorizer.hpp RAGVectorizer.cpp
+    RAGSimilaritySearch.hpp RAGSimilaritySearch.cpp
+    RAGPreprocessor.hpp RAGPreprocessor.cpp
+    EnhancedRAGSimilaritySearch.hpp EnhancedRAGSimilaritySearch.cpp
+    FileChunker.hpp FileChunker.cpp
 )
 
 target_link_libraries(Context
     PUBLIC
     Qt::Core
+    Qt::Sql
     QtCreator::Core
     QtCreator::TextEditor
     QtCreator::Utils

context/ContextManager.cpp

@@ -21,13 +21,12 @@
 
 #include <QFile>
 #include <QFileInfo>
-#include <QJsonObject>
 #include <QTextStream>
 
-#include "GeneralSettings.hpp"
-#include "Logger.hpp"
-#include <texteditor/textdocument.h>
-#include <utils/filepath.h>
+#include <projectexplorer/project.h>
+#include <projectexplorer/projectnodes.h>
+
+#include "FileChunker.hpp"
 
 namespace QodeAssist::Context {
 
@@ -70,33 +69,108 @@ ContentFile ContextManager::createContentFile(const QString &filePath) const
     return contentFile;
 }
 
-ProgrammingLanguage ContextManager::getDocumentLanguage(const QJsonObject &request) const
+bool ContextManager::isInBuildDirectory(const QString &filePath) const
 {
-    QJsonObject params = request["params"].toObject();
-    QJsonObject doc = params["doc"].toObject();
-    QString uri = doc["uri"].toString();
+    static const QStringList buildDirPatterns
+        = {QString(QDir::separator()) + QLatin1String("build") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("Build") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("BUILD") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("debug") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("Debug") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("DEBUG") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("release") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("Release") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("RELEASE") + QDir::separator(),
+           QString(QDir::separator()) + QLatin1String("builds") + QDir::separator()};
 
-    Utils::FilePath filePath = Utils::FilePath::fromString(QUrl(uri).toLocalFile());
-    TextEditor::TextDocument *textDocument = TextEditor::TextDocument::textDocumentForFilePath(
-        filePath);
+    // Нормализуем путь
+    QString normalizedPath = QDir::fromNativeSeparators(filePath);
 
-    if (!textDocument) {
-        LOG_MESSAGE("Error: Document is not available for" + filePath.toString());
-        return Context::ProgrammingLanguage::Unknown;
+    // Проверяем, содержит ли путь паттерны build-директории
+    for (const QString &pattern : buildDirPatterns) {
+        // Сравниваем с нормализованным паттерном
+        QString normalizedPattern = QDir::fromNativeSeparators(pattern);
+        if (normalizedPath.contains(normalizedPattern)) {
+            qDebug() << "Skipping build file:" << filePath;
+            return true;
+        }
     }
 
-    return Context::ProgrammingLanguageUtils::fromMimeType(textDocument->mimeType());
+    return false;
 }
 
-bool ContextManager::isSpecifyCompletion(const QJsonObject &request)
+QStringList ContextManager::getProjectSourceFiles(ProjectExplorer::Project *project) const
 {
-    auto &generalSettings = Settings::generalSettings();
+    QStringList sourceFiles;
+    if (!project)
+        return sourceFiles;
 
-    Context::ProgrammingLanguage documentLanguage = getDocumentLanguage(request);
-    Context::ProgrammingLanguage preset1Language = Context::ProgrammingLanguageUtils::fromString(
-        generalSettings.preset1Language.displayForIndex(generalSettings.preset1Language()));
+    auto projectNode = project->rootProjectNode();
+    if (!projectNode)
+        return sourceFiles;
 
-    return generalSettings.specifyPreset1() && documentLanguage == preset1Language;
+    projectNode->forEachNode(
+        [&sourceFiles, this](ProjectExplorer::FileNode *fileNode) {
+            if (fileNode) {
+                QString filePath = fileNode->filePath().toString();
+                if (shouldProcessFile(filePath) && !isInBuildDirectory(filePath)) {
+                    sourceFiles.append(filePath);
+                }
+            }
+        },
+        nullptr);
+
+    return sourceFiles;
+}
+
+bool ContextManager::shouldProcessFile(const QString &filePath) const
+{
+    static const QStringList supportedExtensions
+        = {"cpp", "hpp", "c", "h", "cc", "hh", "cxx", "hxx", "qml", "js", "py"};
+
+    QFileInfo fileInfo(filePath);
+    return supportedExtensions.contains(fileInfo.suffix().toLower());
+}
+
+void ContextManager::testProjectChunks(
+    ProjectExplorer::Project *project, const FileChunker::ChunkingConfig &config)
+{
+    if (!project) {
+        qDebug() << "No project provided";
+        return;
+    }
+
+    qDebug() << "\nStarting test chunking for project:" << project->displayName();
+
+    // Get source files
+    QStringList sourceFiles = getProjectSourceFiles(project);
+    qDebug() << "Found" << sourceFiles.size() << "source files";
+
+    // Create chunker
+    auto chunker = new FileChunker(config, this);
+
+    // Connect progress and error signals
+    connect(chunker, &FileChunker::progressUpdated, this, [](int processed, int total) {
+        qDebug() << "Progress:" << processed << "/" << total << "files";
+    });
+
+    connect(chunker, &FileChunker::error, this, [](const QString &error) {
+        qDebug() << "Error:" << error;
+    });
+
+    // Start chunking and handle results
+    auto future = chunker->chunkFiles(sourceFiles);
+
+    // Используем QFutureWatcher для обработки результатов
+    auto watcher = new QFutureWatcher<QList<FileChunk>>(this);
+
+    connect(watcher, &QFutureWatcher<QList<FileChunk>>::finished, this, [watcher, chunker]() {
+        // Очистка
+        watcher->deleteLater();
+        chunker->deleteLater();
+    });
+
+    watcher->setFuture(future);
 }
 
 } // namespace QodeAssist::Context

context/ContextManager.hpp

@@ -19,11 +19,16 @@
 
 #pragma once
 
+#include "ContentFile.hpp"
+
 #include <QObject>
 #include <QString>
 
-#include "ContentFile.hpp"
-#include "ProgrammingLanguage.hpp"
+#include "FileChunker.hpp"
+
+namespace ProjectExplorer {
+class Project;
+}
 
 namespace QodeAssist::Context {
 
@@ -33,17 +38,23 @@ class ContextManager : public QObject
 
 public:
     static ContextManager &instance();
+
     QString readFile(const QString &filePath) const;
     QList<ContentFile> getContentFiles(const QStringList &filePaths) const;
-    ProgrammingLanguage getDocumentLanguage(const QJsonObject &request) const;
-    bool isSpecifyCompletion(const QJsonObject &request);
+    QStringList getProjectSourceFiles(ProjectExplorer::Project *project) const;
+
+    void testProjectChunks(
+        ProjectExplorer::Project *project, const FileChunker::ChunkingConfig &config);
 
 private:
     explicit ContextManager(QObject *parent = nullptr);
     ~ContextManager() = default;
     ContextManager(const ContextManager &) = delete;
     ContextManager &operator=(const ContextManager &) = delete;
+
     ContentFile createContentFile(const QString &filePath) const;
+    bool shouldProcessFile(const QString &filePath) const;
+    bool isInBuildDirectory(const QString &filePath) const;
 };
 
 } // namespace QodeAssist::Context

context/DocumentContextReader.cpp

@@ -210,13 +210,14 @@ LLMCore::ContextData DocumentContextReader::prepareContext(int lineNumber, int c
     QString contextAfter = getContextAfter(lineNumber, cursorPosition);
 
     QString fileContext;
-    fileContext.append("\n ").append(getLanguageAndFileInfo());
+    if (Settings::codeCompletionSettings().useFilePathInContext())
+        fileContext.append("\n ").append(getLanguageAndFileInfo());
 
     if (Settings::codeCompletionSettings().useProjectChangesCache())
         fileContext.append("\n ").append(
             ChangesManager::instance().getRecentChangesContext(m_textDocument));
 
-    return {.prefix = contextBefore, .suffix = contextAfter, .fileContext = fileContext};
+    return {contextBefore, contextAfter, fileContext};
 }
 
 QString DocumentContextReader::getContextBefore(int lineNumber, int cursorPosition) const

context/EnhancedRAGSimilaritySearch.cpp

@@ -0,0 +1,265 @@
+#include "EnhancedRAGSimilaritySearch.hpp"
+
+#include <QSet>
+
+namespace QodeAssist::Context {
+
+// Static regex getters
+const QRegularExpression &EnhancedRAGSimilaritySearch::getNamespaceRegex()
+{
+    static const QRegularExpression regex(R"(namespace\s+(?:\w+\s*::\s*)*\w+\s*\{)");
+    return regex;
+}
+
+const QRegularExpression &EnhancedRAGSimilaritySearch::getClassRegex()
+{
+    static const QRegularExpression regex(
+        R"((?:template\s*<[^>]*>\s*)?(?:class|struct)\s+(\w+)\s*(?:final\s*)?(?::\s*(?:public|protected|private)\s+\w+(?:\s*,\s*(?:public|protected|private)\s+\w+)*\s*)?{)");
+    return regex;
+}
+
+const QRegularExpression &EnhancedRAGSimilaritySearch::getFunctionRegex()
+{
+    static const QRegularExpression regex(
+        R"((?:virtual\s+)?(?:static\s+)?(?:inline\s+)?(?:explicit\s+)?(?:constexpr\s+)?(?:[\w:]+\s+)?(?:\w+\s*::\s*)*\w+\s*\([^)]*\)\s*(?:const\s*)?(?:noexcept\s*)?(?:override\s*)?(?:final\s*)?(?:=\s*0\s*)?(?:=\s*default\s*)?(?:=\s*delete\s*)?(?:\s*->.*?)?\s*{)");
+    return regex;
+}
+
+const QRegularExpression &EnhancedRAGSimilaritySearch::getTemplateRegex()
+{
+    static const QRegularExpression regex(R"(template\s*<[^>]*>\s*(?:class|struct|typename)\s+\w+)");
+    return regex;
+}
+
+// Cache getters
+QCache<QString, EnhancedRAGSimilaritySearch::SimilarityScore> &
+EnhancedRAGSimilaritySearch::getScoreCache()
+{
+    static QCache<QString, SimilarityScore> cache(1000); // Cache size of 1000 entries
+    return cache;
+}
+
+QCache<QString, QStringList> &EnhancedRAGSimilaritySearch::getStructureCache()
+{
+    static QCache<QString, QStringList> cache(500); // Cache size of 500 entries
+    return cache;
+}
+
+// Main public interface
+EnhancedRAGSimilaritySearch::SimilarityScore EnhancedRAGSimilaritySearch::calculateSimilarity(
+    const RAGVector &v1, const RAGVector &v2, const QString &code1, const QString &code2)
+{
+    // Generate cache key based on content hashes
+    QString cacheKey = QString("%1_%2").arg(qHash(code1)).arg(qHash(code2));
+
+    // Check cache first
+    auto &scoreCache = getScoreCache();
+    if (auto *cached = scoreCache.object(cacheKey)) {
+        return *cached;
+    }
+
+    // Calculate new similarity score
+    SimilarityScore score = calculateSimilarityInternal(v1, v2, code1, code2);
+
+    // Cache the result
+    scoreCache.insert(cacheKey, new SimilarityScore(score));
+
+    return score;
+}
+
+// Internal implementation
+EnhancedRAGSimilaritySearch::SimilarityScore EnhancedRAGSimilaritySearch::calculateSimilarityInternal(
+    const RAGVector &v1, const RAGVector &v2, const QString &code1, const QString &code2)
+{
+    if (v1.empty() || v2.empty()) {
+        LOG_MESSAGE("Warning: Empty vectors in similarity calculation");
+        return SimilarityScore(0.0f, 0.0f, 0.0f);
+    }
+
+    if (v1.size() != v2.size()) {
+        LOG_MESSAGE(QString("Vector size mismatch: %1 vs %2").arg(v1.size()).arg(v2.size()));
+        return SimilarityScore(0.0f, 0.0f, 0.0f);
+    }
+
+    // Calculate semantic similarity using vector embeddings
+    float semantic_similarity = 0.0f;
+
+#if defined(__SSE__) || defined(_M_X64) || defined(_M_AMD64)
+    if (v1.size() >= 4) { // Use SSE for vectors of 4 or more elements
+        semantic_similarity = calculateCosineSimilaritySSE(v1, v2);
+    } else {
+        semantic_similarity = calculateCosineSimilarity(v1, v2);
+    }
+#else
+    semantic_similarity = calculateCosineSimilarity(v1, v2);
+#endif
+
+    // If semantic similarity is very low, skip structural analysis
+    if (semantic_similarity < 0.0001f) {
+        return SimilarityScore(0.0f, 0.0f, 0.0f);
+    }
+
+    // Calculate structural similarity
+    float structural_similarity = calculateStructuralSimilarity(code1, code2);
+
+    // Calculate combined score with dynamic weights
+    float semantic_weight = 0.7f;
+    const int large_file_threshold = 10000;
+
+    if (code1.size() > large_file_threshold || code2.size() > large_file_threshold) {
+        semantic_weight = 0.8f; // Increase semantic weight for large files
+    }
+
+    float combined_score = semantic_weight * semantic_similarity
+                           + (1.0f - semantic_weight) * structural_similarity;
+
+    return SimilarityScore(semantic_similarity, structural_similarity, combined_score);
+}
+
+float EnhancedRAGSimilaritySearch::calculateCosineSimilarity(const RAGVector &v1, const RAGVector &v2)
+{
+    float dotProduct = 0.0f;
+    float norm1 = 0.0f;
+    float norm2 = 0.0f;
+
+    for (size_t i = 0; i < v1.size(); ++i) {
+        dotProduct += v1[i] * v2[i];
+        norm1 += v1[i] * v1[i];
+        norm2 += v2[i] * v2[i];
+    }
+
+    norm1 = std::sqrt(norm1);
+    norm2 = std::sqrt(norm2);
+
+    if (norm1 == 0.0f || norm2 == 0.0f) {
+        return 0.0f;
+    }
+
+    return dotProduct / (norm1 * norm2);
+}
+
+#if defined(__SSE__) || defined(_M_X64) || defined(_M_AMD64)
+float EnhancedRAGSimilaritySearch::calculateCosineSimilaritySSE(
+    const RAGVector &v1, const RAGVector &v2)
+{
+    const float *p1 = v1.data();
+    const float *p2 = v2.data();
+    const size_t size = v1.size();
+    const size_t alignedSize = size & ~3ULL; // Round down to multiple of 4
+
+    __m128 sum = _mm_setzero_ps();
+    __m128 norm1 = _mm_setzero_ps();
+    __m128 norm2 = _mm_setzero_ps();
+
+    // Process 4 elements at a time using SSE
+    for (size_t i = 0; i < alignedSize; i += 4) {
+        __m128 v1_vec = _mm_loadu_ps(p1 + i); // Use unaligned load for safety
+        __m128 v2_vec = _mm_loadu_ps(p2 + i);
+
+        sum = _mm_add_ps(sum, _mm_mul_ps(v1_vec, v2_vec));
+        norm1 = _mm_add_ps(norm1, _mm_mul_ps(v1_vec, v1_vec));
+        norm2 = _mm_add_ps(norm2, _mm_mul_ps(v2_vec, v2_vec));
+    }
+
+    float dotProduct = horizontalSum(sum);
+    float n1 = std::sqrt(horizontalSum(norm1));
+    float n2 = std::sqrt(horizontalSum(norm2));
+
+    // Process remaining elements
+    for (size_t i = alignedSize; i < size; ++i) {
+        dotProduct += v1[i] * v2[i];
+        n1 += v1[i] * v1[i];
+        n2 += v2[i] * v2[i];
+    }
+
+    if (n1 == 0.0f || n2 == 0.0f) {
+        return 0.0f;
+    }
+
+    return dotProduct / (std::sqrt(n1) * std::sqrt(n2));
+}
+
+float EnhancedRAGSimilaritySearch::horizontalSum(__m128 x)
+{
+    __m128 shuf = _mm_shuffle_ps(x, x, _MM_SHUFFLE(2, 3, 0, 1));
+    __m128 sums = _mm_add_ps(x, shuf);
+    shuf = _mm_movehl_ps(shuf, sums);
+    sums = _mm_add_ss(sums, shuf);
+    return _mm_cvtss_f32(sums);
+}
+#endif
+
+float EnhancedRAGSimilaritySearch::calculateStructuralSimilarity(
+    const QString &code1, const QString &code2)
+{
+    QStringList structures1 = extractStructures(code1);
+    QStringList structures2 = extractStructures(code2);
+
+    return calculateJaccardSimilarity(structures1, structures2);
+}
+
+QStringList EnhancedRAGSimilaritySearch::extractStructures(const QString &code)
+{
+    // Check cache first
+    auto &structureCache = getStructureCache();
+    QString cacheKey = QString::number(qHash(code));
+
+    if (auto *cached = structureCache.object(cacheKey)) {
+        return *cached;
+    }
+
+    QStringList structures;
+    structures.reserve(100); // Reserve space for typical file
+
+    // Extract namespaces
+    auto namespaceMatches = getNamespaceRegex().globalMatch(code);
+    while (namespaceMatches.hasNext()) {
+        structures.append(namespaceMatches.next().captured().trimmed());
+    }
+
+    // Extract classes
+    auto classMatches = getClassRegex().globalMatch(code);
+    while (classMatches.hasNext()) {
+        structures.append(classMatches.next().captured().trimmed());
+    }
+
+    // Extract functions
+    auto functionMatches = getFunctionRegex().globalMatch(code);
+    while (functionMatches.hasNext()) {
+        structures.append(functionMatches.next().captured().trimmed());
+    }
+
+    // Extract templates
+    auto templateMatches = getTemplateRegex().globalMatch(code);
+    while (templateMatches.hasNext()) {
+        structures.append(templateMatches.next().captured().trimmed());
+    }
+
+    // Cache the result
+    structureCache.insert(cacheKey, new QStringList(structures));
+
+    return structures;
+}
+
+float EnhancedRAGSimilaritySearch::calculateJaccardSimilarity(
+    const QStringList &set1, const QStringList &set2)
+{
+    if (set1.isEmpty() && set2.isEmpty()) {
+        return 1.0f; // Пустые множества считаем идентичными
+    }
+    if (set1.isEmpty() || set2.isEmpty()) {
+        return 0.0f;
+    }
+
+    QSet<QString> set1Unique = QSet<QString>(set1.begin(), set1.end());
+    QSet<QString> set2Unique = QSet<QString>(set2.begin(), set2.end());
+
+    QSet<QString> intersection = set1Unique;
+    intersection.intersect(set2Unique);
+
+    QSet<QString> union_set = set1Unique;
+    union_set.unite(set2Unique);
+
+    return static_cast<float>(intersection.size()) / union_set.size();
+}
+} // namespace QodeAssist::Context

context/EnhancedRAGSimilaritySearch.hpp

@@ -0,0 +1,74 @@
+#pragma once
+
+#include <QCache>
+#include <QHash>
+#include <QRegularExpression>
+#include <QString>
+#include <QStringList>
+#include <QtGlobal>
+
+#include <algorithm>
+#include <cmath>
+#include <optional>
+#include <vector>
+
+#if defined(__SSE__) || defined(_M_X64) || defined(_M_AMD64)
+#include <emmintrin.h>
+#include <xmmintrin.h>
+#endif
+
+#include "RAGData.hpp"
+#include "logger/Logger.hpp"
+
+namespace QodeAssist::Context {
+
+class EnhancedRAGSimilaritySearch
+{
+public:
+    struct SimilarityScore
+    {
+        float semantic_similarity{0.0f};
+        float structural_similarity{0.0f};
+        float combined_score{0.0f};
+
+        SimilarityScore() = default;
+        SimilarityScore(float sem, float str, float comb)
+            : semantic_similarity(sem)
+            , structural_similarity(str)
+            , combined_score(comb)
+        {}
+    };
+
+    static SimilarityScore calculateSimilarity(
+        const RAGVector &v1, const RAGVector &v2, const QString &code1, const QString &code2);
+
+private:
+    static SimilarityScore calculateSimilarityInternal(
+        const RAGVector &v1, const RAGVector &v2, const QString &code1, const QString &code2);
+
+    static float calculateCosineSimilarity(const RAGVector &v1, const RAGVector &v2);
+
+#if defined(__SSE__) || defined(_M_X64) || defined(_M_AMD64)
+    static float calculateCosineSimilaritySSE(const RAGVector &v1, const RAGVector &v2);
+    static float horizontalSum(__m128 x);
+#endif
+
+    static float calculateStructuralSimilarity(const QString &code1, const QString &code2);
+    static QStringList extractStructures(const QString &code);
+    static float calculateJaccardSimilarity(const QStringList &set1, const QStringList &set2);
+
+    static const QRegularExpression &getNamespaceRegex();
+    static const QRegularExpression &getClassRegex();
+    static const QRegularExpression &getFunctionRegex();
+    static const QRegularExpression &getTemplateRegex();
+
+    // Cache for similarity scores
+    static QCache<QString, SimilarityScore> &getScoreCache();
+
+    // Cache for extracted structures
+    static QCache<QString, QStringList> &getStructureCache();
+
+    EnhancedRAGSimilaritySearch() = delete; // Prevent instantiation
+};
+
+} // namespace QodeAssist::Context

context/FileChunker.cpp

@@ -0,0 +1,198 @@
+// FileChunker.cpp
+#include "FileChunker.hpp"
+
+#include <coreplugin/idocument.h>
+#include <texteditor/syntaxhighlighter.h>
+#include <texteditor/textdocument.h>
+#include <texteditor/texteditorconstants.h>
+
+#include <QFutureWatcher>
+#include <QTimer>
+
+namespace QodeAssist::Context {
+
+FileChunker::FileChunker(QObject *parent)
+    : QObject(parent)
+{}
+
+FileChunker::FileChunker(const ChunkingConfig &config, QObject *parent)
+    : QObject(parent)
+    , m_config(config)
+{}
+
+QFuture<QList<FileChunk>> FileChunker::chunkFiles(const QStringList &filePaths)
+{
+    qDebug() << "\nStarting chunking process for" << filePaths.size() << "files";
+    qDebug() << "Configuration:"
+             << "\n  Max lines per chunk:" << m_config.maxLinesPerChunk
+             << "\n  Overlap lines:" << m_config.overlapLines
+             << "\n  Skip empty lines:" << m_config.skipEmptyLines
+             << "\n  Preserve functions:" << m_config.preserveFunctions
+             << "\n  Preserve classes:" << m_config.preserveClasses
+             << "\n  Batch size:" << m_config.batchSize;
+
+    auto promise = std::make_shared<QPromise<QList<FileChunk>>>();
+    promise->start();
+
+    if (filePaths.isEmpty()) {
+        qDebug() << "No files to process";
+        promise->addResult({});
+        promise->finish();
+        return promise->future();
+    }
+
+    processNextBatch(promise, filePaths, 0);
+    return promise->future();
+}
+
+void FileChunker::processNextBatch(
+    std::shared_ptr<QPromise<QList<FileChunk>>> promise, const QStringList &files, int startIndex)
+{
+    if (startIndex >= files.size()) {
+        emit chunkingComplete();
+        promise->finish();
+        return;
+    }
+
+    int endIndex = qMin(startIndex + m_config.batchSize, files.size());
+    QList<FileChunk> batchChunks;
+
+    for (int i = startIndex; i < endIndex; ++i) {
+        try {
+            auto chunks = processFile(files[i]);
+            batchChunks.append(chunks);
+        } catch (const std::exception &e) {
+            emit error(QString("Error processing file %1: %2").arg(files[i], e.what()));
+        }
+        emit progressUpdated(i + 1, files.size());
+    }
+
+    promise->addResult(batchChunks);
+
+    // Планируем обработку следующего батча
+    QTimer::singleShot(0, this, [this, promise, files, endIndex]() {
+        processNextBatch(promise, files, endIndex);
+    });
+}
+
+QList<FileChunk> FileChunker::processFile(const QString &filePath)
+{
+    qDebug() << "\nProcessing file:" << filePath;
+
+    auto document = new TextEditor::TextDocument;
+    auto filePathObj = Utils::FilePath::fromString(filePath);
+    auto result = document->open(&m_error, filePathObj, filePathObj);
+    if (result != Core::IDocument::OpenResult::Success) {
+        qDebug() << "Failed to open document:" << filePath << "-" << m_error;
+        emit error(QString("Failed to open document: %1 - %2").arg(filePath, m_error));
+        delete document;
+        return {};
+    }
+
+    qDebug() << "Document opened successfully. Line count:" << document->document()->blockCount();
+
+    auto chunks = createChunksForDocument(document);
+    qDebug() << "Created" << chunks.size() << "chunks for file";
+
+    delete document;
+    return chunks;
+}
+
+QList<FileChunk> FileChunker::createChunksForDocument(TextEditor::TextDocument *document)
+{
+    QList<FileChunk> chunks;
+    QString filePath = document->filePath().toString();
+    qDebug() << "\nCreating chunks for document:" << filePath << "\nConfiguration:"
+             << "\n  Max lines per chunk:" << m_config.maxLinesPerChunk
+             << "\n  Min lines per chunk:" << m_config.minLinesPerChunk
+             << "\n  Overlap lines:" << m_config.overlapLines;
+    // Если файл меньше минимального размера чанка, создаем один чанк
+    if (document->document()->blockCount() <= m_config.minLinesPerChunk) {
+        FileChunk chunk;
+        chunk.filePath = filePath;
+        chunk.startLine = 0;
+        chunk.endLine = document->document()->blockCount() - 1;
+        chunk.createdAt = QDateTime::currentDateTime();
+        chunk.updatedAt = chunk.createdAt;
+
+        QString content;
+        QTextBlock block = document->document()->firstBlock();
+        while (block.isValid()) {
+            content += block.text() + "\n";
+            block = block.next();
+        }
+        chunk.content = content;
+
+        qDebug() << "File is smaller than minimum chunk size. Creating single chunk:"
+                 << "\n  Lines:" << chunk.lineCount() << "\n  Content size:" << chunk.content.size()
+                 << "bytes";
+
+        chunks.append(chunk);
+        return chunks;
+    }
+
+    // Для больших файлов создаем чанки фиксированного размера с перекрытием
+    int currentStartLine = 0;
+    int lineCount = 0;
+    QString content;
+    QTextBlock block = document->document()->firstBlock();
+
+    while (block.isValid()) {
+        content += block.text() + "\n";
+        lineCount++;
+
+        // Если достигли размера чанка или это последний блок
+        if (lineCount >= m_config.maxLinesPerChunk || !block.next().isValid()) {
+            FileChunk chunk;
+            chunk.filePath = filePath;
+            chunk.startLine = currentStartLine;
+            chunk.endLine = currentStartLine + lineCount - 1;
+            chunk.content = content;
+            chunk.createdAt = QDateTime::currentDateTime();
+            chunk.updatedAt = chunk.createdAt;
+
+            qDebug() << "Creating chunk:"
+                     << "\n  Start line:" << chunk.startLine << "\n  End line:" << chunk.endLine
+                     << "\n  Lines:" << chunk.lineCount()
+                     << "\n  Content size:" << chunk.content.size() << "bytes";
+
+            chunks.append(chunk);
+
+            // Начинаем новый чанк с учетом перекрытия
+            if (block.next().isValid()) {
+                // Отступаем назад на размер перекрытия
+                int overlapLines = qMin(m_config.overlapLines, lineCount);
+                currentStartLine = chunk.endLine - overlapLines + 1;
+
+                // Сбрасываем контент, но добавляем перекрывающиеся строки
+                content.clear();
+                QTextBlock overlapBlock = document->document()->findBlockByLineNumber(
+                    currentStartLine);
+                while (overlapBlock.isValid() && overlapBlock.blockNumber() <= chunk.endLine) {
+                    content += overlapBlock.text() + "\n";
+                    overlapBlock = overlapBlock.next();
+                }
+                lineCount = overlapLines;
+            }
+        }
+
+        block = block.next();
+    }
+
+    qDebug() << "Finished creating chunks for file:" << filePath
+             << "\nTotal chunks:" << chunks.size();
+
+    return chunks;
+}
+
+void FileChunker::setConfig(const ChunkingConfig &config)
+{
+    m_config = config;
+}
+
+FileChunker::ChunkingConfig FileChunker::config() const
+{
+    return m_config;
+}
+
+} // namespace QodeAssist::Context

context/FileChunker.hpp

@@ -0,0 +1,68 @@
+// FileChunker.hpp
+#pragma once
+
+#include <texteditor/textdocument.h>
+#include <QDateTime>
+#include <QFuture>
+#include <QString>
+
+namespace QodeAssist::Context {
+
+struct FileChunk
+{
+    QString filePath;    // Path to the source file
+    int startLine;       // Starting line of the chunk
+    int endLine;         // Ending line of the chunk
+    QDateTime createdAt; // When the chunk was created
+    QDateTime updatedAt; // When the chunk was last updated
+    QString content;     // Content of the chunk
+
+    // Helper methods
+    int lineCount() const { return endLine - startLine + 1; }
+    bool isValid() const { return !filePath.isEmpty() && startLine >= 0 && endLine >= startLine; }
+};
+
+class FileChunker : public QObject
+{
+    Q_OBJECT
+
+public:
+    struct ChunkingConfig
+    {
+        int maxLinesPerChunk = 80;
+        int minLinesPerChunk = 40;
+        int overlapLines = 20;
+        bool skipEmptyLines = true;
+        bool preserveFunctions = true;
+        bool preserveClasses = true;
+        int batchSize = 10;
+    };
+
+    explicit FileChunker(QObject *parent = nullptr);
+    explicit FileChunker(const ChunkingConfig &config, QObject *parent = nullptr);
+
+    // Main chunking method
+    QFuture<QList<FileChunk>> chunkFiles(const QStringList &filePaths);
+
+    // Configuration
+    void setConfig(const ChunkingConfig &config);
+    ChunkingConfig config() const;
+
+signals:
+    void progressUpdated(int processedFiles, int totalFiles);
+    void chunkingComplete();
+    void error(const QString &errorMessage);
+
+private:
+    QList<FileChunk> processFile(const QString &filePath);
+    QList<FileChunk> createChunksForDocument(TextEditor::TextDocument *document);
+    void processNextBatch(
+        std::shared_ptr<QPromise<QList<FileChunk>>> promise,
+        const QStringList &files,
+        int startIndex);
+
+    ChunkingConfig m_config;
+    QString m_error;
+};
+
+} // namespace QodeAssist::Context

context/ProgrammingLanguage.hpp

@@ -24,8 +24,8 @@
 namespace QodeAssist::Context {
 
 enum class ProgrammingLanguage {
-    QML, // QML/JavaScript
-    Cpp, // C/C++
+    QML,
+    Cpp,
     Python,
     Unknown,
 };

context/RAGData.hpp

@@ -0,0 +1,7 @@
+#pragma once
+
+#include <vector>
+
+namespace QodeAssist::Context {
+    using RAGVector = std::vector<float>;
+}

context/RAGManager.cpp

@@ -0,0 +1,443 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "RAGManager.hpp"
+#include "EnhancedRAGSimilaritySearch.hpp"
+#include "RAGPreprocessor.hpp"
+#include "RAGSimilaritySearch.hpp"
+#include "logger/Logger.hpp"
+
+#include <coreplugin/icore.h>
+#include <projectexplorer/project.h>
+#include <QFile>
+#include <QtConcurrent>
+
+namespace QodeAssist::Context {
+
+RAGManager &RAGManager::instance()
+{
+    static RAGManager manager;
+    return manager;
+}
+
+RAGManager::RAGManager(QObject *parent)
+    : QObject(parent)
+    , m_vectorizer(std::make_unique<RAGVectorizer>())
+{}
+
+RAGManager::~RAGManager() {}
+
+QString RAGManager::getStoragePath(ProjectExplorer::Project *project) const
+{
+    return QString("%1/qodeassist/%2/rag/vectors.db")
+        .arg(Core::ICore::userResourcePath().toString(), project->displayName());
+}
+
+std::optional<QString> RAGManager::loadFileContent(const QString &filePath)
+{
+    QFile file(filePath);
+    if (!file.open(QIODevice::ReadOnly | QIODevice::Text)) {
+        qDebug() << "ERROR: Failed to open file for reading:" << filePath
+                 << "Error:" << file.errorString();
+        return std::nullopt;
+    }
+
+    QFileInfo fileInfo(filePath);
+    qDebug() << "Loading content from file:" << fileInfo.fileName() << "Size:" << fileInfo.size()
+             << "bytes";
+
+    QString content = QString::fromUtf8(file.readAll());
+    if (content.isEmpty()) {
+        qDebug() << "WARNING: Empty content read from file:" << filePath;
+    }
+    return content;
+}
+
+void RAGManager::ensureStorageForProject(ProjectExplorer::Project *project) const
+{
+    qDebug() << "Ensuring storage for project:" << project->displayName();
+
+    if (m_currentProject == project && m_currentStorage) {
+        qDebug() << "Using existing storage";
+        return;
+    }
+
+    qDebug() << "Creating new storage";
+    m_currentStorage.reset();
+    m_currentProject = project;
+
+    if (project) {
+        QString storagePath = getStoragePath(project);
+        qDebug() << "Storage path:" << storagePath;
+
+        StorageOptions options;
+        m_currentStorage = std::make_unique<RAGStorage>(storagePath, options);
+
+        qDebug() << "Initializing storage...";
+        if (!m_currentStorage->init()) {
+            qDebug() << "Failed to initialize storage";
+            m_currentStorage.reset();
+            return;
+        }
+        qDebug() << "Storage initialized successfully";
+    }
+}
+
+QFuture<void> RAGManager::processProjectFiles(
+    ProjectExplorer::Project *project,
+    const QStringList &filePaths,
+    const FileChunker::ChunkingConfig &config)
+{
+    qDebug() << "\nStarting batch processing of" << filePaths.size()
+             << "files for project:" << project->displayName();
+
+    auto promise = std::make_shared<QPromise<void>>();
+    promise->start();
+
+    qDebug() << "Initializing storage...";
+    ensureStorageForProject(project);
+
+    if (!m_currentStorage) {
+        qDebug() << "Failed to initialize storage for project:" << project->displayName();
+        promise->finish();
+        return promise->future();
+    }
+    qDebug() << "Storage initialized successfully";
+
+    qDebug() << "Checking files for processing...";
+    QSet<QString> uniqueFiles;
+    for (const QString &filePath : filePaths) {
+        qDebug() << "Checking file:" << filePath;
+        if (isFileStorageOutdated(project, filePath)) {
+            qDebug() << "File needs processing:" << filePath;
+            uniqueFiles.insert(filePath);
+        }
+    }
+
+    QStringList filesToProcess = uniqueFiles.values();
+
+    if (filesToProcess.isEmpty()) {
+        qDebug() << "No files need processing";
+        emit vectorizationFinished();
+        promise->finish();
+        return promise->future();
+    }
+
+    qDebug() << "Starting to process" << filesToProcess.size() << "files";
+    const int batchSize = 10;
+    processNextFileBatch(promise, project, filesToProcess, config, 0, batchSize);
+
+    return promise->future();
+}
+
+void RAGManager::processNextFileBatch(
+    std::shared_ptr<QPromise<void>> promise,
+    ProjectExplorer::Project *project,
+    const QStringList &files,
+    const FileChunker::ChunkingConfig &config,
+    int startIndex,
+    int batchSize)
+{
+    if (startIndex >= files.size()) {
+        qDebug() << "All batches processed successfully";
+        emit vectorizationFinished();
+        promise->finish();
+        return;
+    }
+
+    int endIndex = qMin(startIndex + batchSize, files.size());
+    auto currentBatch = files.mid(startIndex, endIndex - startIndex);
+
+    qDebug() << "\nProcessing batch" << (startIndex / batchSize + 1) << "(" << currentBatch.size()
+             << "files)"
+             << "\nProgress:" << startIndex << "to" << endIndex << "of" << files.size();
+
+    for (const QString &filePath : currentBatch) {
+        qDebug() << "Starting processing file:" << filePath;
+        auto future = processFileWithChunks(project, filePath, config);
+        auto watcher = new QFutureWatcher<bool>;
+        watcher->setFuture(future);
+
+        connect(
+            watcher,
+            &QFutureWatcher<bool>::finished,
+            this,
+            [this,
+             watcher,
+             promise,
+             project,
+             files,
+             startIndex,
+             endIndex,
+             batchSize,
+             config,
+             filePath]() {
+                bool success = watcher->result();
+                qDebug() << "File processed:" << filePath << "success:" << success;
+
+                bool isLastFileInBatch = (filePath == files[endIndex - 1]);
+                if (isLastFileInBatch) {
+                    qDebug() << "Batch completed, moving to next batch";
+                    emit vectorizationProgress(endIndex, files.size());
+                    processNextFileBatch(promise, project, files, config, endIndex, batchSize);
+                }
+
+                watcher->deleteLater();
+            });
+    }
+}
+
+QFuture<bool> RAGManager::processFileWithChunks(
+    ProjectExplorer::Project *project,
+    const QString &filePath,
+    const FileChunker::ChunkingConfig &config)
+{
+    auto promise = std::make_shared<QPromise<bool>>();
+    promise->start();
+
+    ensureStorageForProject(project);
+    if (!m_currentStorage) {
+        qDebug() << "Storage not initialized for file:" << filePath;
+        promise->addResult(false);
+        promise->finish();
+        return promise->future();
+    }
+
+    auto fileContent = loadFileContent(filePath);
+    if (!fileContent) {
+        qDebug() << "Failed to load content for file:" << filePath;
+        promise->addResult(false);
+        promise->finish();
+        return promise->future();
+    }
+
+    qDebug() << "Creating chunks for file:" << filePath;
+    auto chunksFuture = m_chunker.chunkFiles({filePath});
+    auto chunks = chunksFuture.result();
+
+    if (chunks.isEmpty()) {
+        qDebug() << "No chunks created for file:" << filePath;
+        promise->addResult(false);
+        promise->finish();
+        return promise->future();
+    }
+
+    qDebug() << "Created" << chunks.size() << "chunks for file:" << filePath;
+
+    // Преобразуем FileChunk в FileChunkData
+    QList<FileChunkData> chunkData;
+    for (const auto &chunk : chunks) {
+        FileChunkData data;
+        data.filePath = chunk.filePath;
+        data.startLine = chunk.startLine;
+        data.endLine = chunk.endLine;
+        data.content = chunk.content;
+        chunkData.append(data);
+    }
+
+    qDebug() << "Deleting old chunks for file:" << filePath;
+    m_currentStorage->deleteChunksForFile(filePath);
+
+    auto vectorizeFuture = vectorizeAndStoreChunks(filePath, chunkData);
+    auto watcher = new QFutureWatcher<void>;
+    watcher->setFuture(vectorizeFuture);
+
+    connect(watcher, &QFutureWatcher<void>::finished, this, [promise, watcher, filePath]() {
+        qDebug() << "Completed processing file:" << filePath;
+        promise->addResult(true);
+        promise->finish();
+        watcher->deleteLater();
+    });
+
+    return promise->future();
+}
+
+QFuture<void> RAGManager::vectorizeAndStoreChunks(
+    const QString &filePath, const QList<FileChunkData> &chunks)
+{
+    qDebug() << "Vectorizing and storing" << chunks.size() << "chunks for file:" << filePath;
+
+    auto promise = std::make_shared<QPromise<void>>();
+    promise->start();
+
+    // Обрабатываем чанки последовательно
+    processNextChunk(promise, chunks, 0);
+
+    return promise->future();
+}
+
+void RAGManager::processNextChunk(
+    std::shared_ptr<QPromise<void>> promise, const QList<FileChunkData> &chunks, int currentIndex)
+{
+    if (currentIndex >= chunks.size()) {
+        promise->finish();
+        return;
+    }
+
+    const auto &chunk = chunks[currentIndex];
+    QString processedContent = RAGPreprocessor::preprocessCode(chunk.content);
+    qDebug() << "Processing chunk" << currentIndex + 1 << "of" << chunks.size();
+
+    auto vectorFuture = m_vectorizer->vectorizeText(processedContent);
+    auto watcher = new QFutureWatcher<RAGVector>;
+    watcher->setFuture(vectorFuture);
+
+    connect(
+        watcher,
+        &QFutureWatcher<RAGVector>::finished,
+        this,
+        [this, watcher, promise, chunks, currentIndex, chunk]() {
+            auto vector = watcher->result();
+
+            if (!vector.empty()) {
+                qDebug() << "Storing vector and chunk for file:" << chunk.filePath;
+                bool vectorStored = m_currentStorage->storeVector(chunk.filePath, vector);
+                bool chunkStored = m_currentStorage->storeChunk(chunk);
+                qDebug() << "Storage results - Vector:" << vectorStored << "Chunk:" << chunkStored;
+            } else {
+                qDebug() << "Failed to vectorize chunk content";
+            }
+
+            processNextChunk(promise, chunks, currentIndex + 1);
+
+            watcher->deleteLater();
+        });
+}
+
+QFuture<QList<RAGManager::ChunkSearchResult>> RAGManager::findRelevantChunks(
+    const QString &query, ProjectExplorer::Project *project, int topK)
+{
+    auto promise = std::make_shared<QPromise<QList<ChunkSearchResult>>>();
+    promise->start();
+
+    ensureStorageForProject(project);
+    if (!m_currentStorage) {
+        qDebug() << "Storage not initialized for project:" << project->displayName();
+        promise->addResult({});
+        promise->finish();
+        return promise->future();
+    }
+
+    QString processedQuery = RAGPreprocessor::preprocessCode(query);
+
+    auto vectorFuture = m_vectorizer->vectorizeText(processedQuery);
+    vectorFuture.then([this, promise, project, processedQuery, topK](const RAGVector &queryVector) {
+        if (queryVector.empty()) {
+            qDebug() << "Failed to vectorize query";
+            promise->addResult({});
+            promise->finish();
+            return;
+        }
+
+        auto files = m_currentStorage->getFilesWithChunks();
+        QList<FileChunkData> allChunks;
+
+        for (const auto &filePath : files) {
+            auto fileChunks = m_currentStorage->getChunksForFile(filePath);
+            allChunks.append(fileChunks);
+        }
+
+        auto results = rankChunks(queryVector, processedQuery, allChunks);
+
+        if (results.size() > topK) {
+            results = results.mid(0, topK);
+        }
+
+        qDebug() << "Found" << results.size() << "relevant chunks";
+        promise->addResult(results);
+        promise->finish();
+
+        closeStorage();
+    });
+
+    return promise->future();
+}
+
+QList<RAGManager::ChunkSearchResult> RAGManager::rankChunks(
+    const RAGVector &queryVector, const QString &queryText, const QList<FileChunkData> &chunks)
+{
+    QList<ChunkSearchResult> results;
+    results.reserve(chunks.size());
+
+    for (const auto &chunk : chunks) {
+        auto chunkVector = m_currentStorage->getVector(chunk.filePath);
+        if (!chunkVector.has_value()) {
+            continue;
+        }
+
+        QString processedChunk = RAGPreprocessor::preprocessCode(chunk.content);
+
+        auto similarity = EnhancedRAGSimilaritySearch::calculateSimilarity(
+            queryVector, chunkVector.value(), queryText, processedChunk);
+
+        results.append(ChunkSearchResult{
+            chunk.filePath,
+            chunk.startLine,
+            chunk.endLine,
+            chunk.content,
+            similarity.semantic_similarity,
+            similarity.structural_similarity,
+            similarity.combined_score});
+    }
+
+    std::sort(results.begin(), results.end());
+
+    return results;
+}
+
+QStringList RAGManager::getStoredFiles(ProjectExplorer::Project *project) const
+{
+    ensureStorageForProject(project);
+    if (!m_currentStorage) {
+        return {};
+    }
+    return m_currentStorage->getAllFiles();
+}
+
+bool RAGManager::isFileStorageOutdated(
+    ProjectExplorer::Project *project, const QString &filePath) const
+{
+    ensureStorageForProject(project);
+    if (!m_currentStorage) {
+        return true;
+    }
+    return m_currentStorage->needsUpdate(filePath);
+}
+
+std::optional<RAGVector> RAGManager::loadVectorFromStorage(
+    ProjectExplorer::Project *project, const QString &filePath)
+{
+    ensureStorageForProject(project);
+    if (!m_currentStorage) {
+        return std::nullopt;
+    }
+    return m_currentStorage->getVector(filePath);
+}
+
+void RAGManager::closeStorage()
+{
+    qDebug() << "Closing storage...";
+    if (m_currentStorage) {
+        m_currentStorage.reset();
+        m_currentProject = nullptr;
+        qDebug() << "Storage closed";
+    }
+}
+
+} // namespace QodeAssist::Context

context/RAGManager.hpp

@@ -0,0 +1,119 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <memory>
+#include <optional>
+#include <QFuture>
+#include <QObject>
+
+#include "FileChunker.hpp"
+#include "RAGData.hpp"
+#include "RAGStorage.hpp"
+#include "RAGVectorizer.hpp"
+
+namespace ProjectExplorer {
+class Project;
+}
+
+namespace QodeAssist::Context {
+
+class RAGManager : public QObject
+{
+    Q_OBJECT
+
+public:
+    struct ChunkSearchResult
+    {
+        QString filePath;
+        int startLine;
+        int endLine;
+        QString content;
+        float semanticScore;
+        float structuralScore;
+        float combinedScore;
+
+        bool operator<(const ChunkSearchResult &other) const
+        {
+            return combinedScore > other.combinedScore;
+        }
+    };
+
+    static RAGManager &instance();
+
+    QFuture<void> processProjectFiles(
+        ProjectExplorer::Project *project,
+        const QStringList &filePaths,
+        const FileChunker::ChunkingConfig &config = FileChunker::ChunkingConfig());
+
+    QFuture<QList<ChunkSearchResult>> findRelevantChunks(
+        const QString &query, ProjectExplorer::Project *project, int topK = 5);
+
+    QStringList getStoredFiles(ProjectExplorer::Project *project) const;
+    bool isFileStorageOutdated(ProjectExplorer::Project *project, const QString &filePath) const;
+
+    void processNextChunk(
+        std::shared_ptr<QPromise<void>> promise,
+        const QList<FileChunkData> &chunks,
+        int currentIndex);
+    void closeStorage();
+signals:
+    void vectorizationProgress(int processed, int total);
+    void vectorizationFinished();
+
+private:
+    explicit RAGManager(QObject *parent = nullptr);
+    ~RAGManager();
+    RAGManager(const RAGManager &) = delete;
+    RAGManager &operator=(const RAGManager &) = delete;
+
+    QString getStoragePath(ProjectExplorer::Project *project) const;
+    void ensureStorageForProject(ProjectExplorer::Project *project) const;
+    std::optional<QString> loadFileContent(const QString &filePath);
+    std::optional<RAGVector> loadVectorFromStorage(
+        ProjectExplorer::Project *project, const QString &filePath);
+
+    void processNextFileBatch(
+        std::shared_ptr<QPromise<void>> promise,
+        ProjectExplorer::Project *project,
+        const QStringList &files,
+        const FileChunker::ChunkingConfig &config,
+        int startIndex,
+        int batchSize);
+
+    QFuture<bool> processFileWithChunks(
+        ProjectExplorer::Project *project,
+        const QString &filePath,
+        const FileChunker::ChunkingConfig &config);
+
+    QFuture<void> vectorizeAndStoreChunks(
+        const QString &filePath, const QList<FileChunkData> &chunks);
+
+    QList<ChunkSearchResult> rankChunks(
+        const RAGVector &queryVector, const QString &queryText, const QList<FileChunkData> &chunks);
+
+private:
+    mutable std::unique_ptr<RAGVectorizer> m_vectorizer;
+    mutable std::unique_ptr<RAGStorage> m_currentStorage;
+    mutable ProjectExplorer::Project *m_currentProject{nullptr};
+    FileChunker m_chunker;
+};
+
+} // namespace QodeAssist::Context

context/RAGPreprocessor.cpp

@@ -0,0 +1,2 @@
+#include "RAGPreprocessor.hpp"
+

context/RAGPreprocessor.hpp

@@ -0,0 +1,64 @@
+#include <QRegularExpression>
+#include <QString>
+
+#include "Logger.hpp"
+
+namespace QodeAssist::Context {
+
+class RAGPreprocessor
+{
+public:
+    static const QRegularExpression &getLicenseRegex()
+    {
+        static const QRegularExpression regex(
+            R"((/\*[^*]*\*+(?:[^/*][^*]*\*+)*/)|//[^\n]*(?:\n|$))",
+            QRegularExpression::MultilineOption);
+        return regex;
+    }
+
+    static const QRegularExpression &getClassRegex()
+    {
+        static const QRegularExpression regex(
+            R"((?:template\s*<[^>]*>\s*)?(?:class|struct)\s+(\w+)\s*(?:final\s*)?(?::\s*(?:public|protected|private)\s+\w+(?:\s*,\s*(?:public|protected|private)\s+\w+)*\s*)?{)");
+        return regex;
+    }
+
+    static QString preprocessCode(const QString &code)
+    {
+        if (code.isEmpty()) {
+            return QString();
+        }
+
+        try {
+            QStringList lines = code.split('\n', Qt::SkipEmptyParts);
+            return processLines(lines);
+        } catch (const std::exception &e) {
+            LOG_MESSAGE(QString("Error preprocessing code: %1").arg(e.what()));
+            return code;
+        }
+    }
+
+private:
+    static QString processLines(const QStringList &lines)
+    {
+        const int estimatedAvgLength = 80;
+        QString result;
+        result.reserve(lines.size() * estimatedAvgLength);
+
+        for (const QString &line : lines) {
+            const QString trimmed = line.trimmed();
+            if (!trimmed.isEmpty()) {
+                result += trimmed;
+                result += QLatin1Char('\n');
+            }
+        }
+
+        if (result.endsWith('\n')) {
+            result.chop(1);
+        }
+
+        return result;
+    }
+};
+
+} // namespace QodeAssist::Context

context/RAGSimilaritySearch.cpp

@@ -0,0 +1,67 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "RAGSimilaritySearch.hpp"
+#include "logger/Logger.hpp"
+
+#include <cmath>
+
+namespace QodeAssist::Context {
+
+float RAGSimilaritySearch::l2Distance(const RAGVector &v1, const RAGVector &v2)
+{
+    if (v1.size() != v2.size()) {
+        LOG_MESSAGE(QString("Vector size mismatch: %1 vs %2").arg(v1.size()).arg(v2.size()));
+        return std::numeric_limits<float>::max();
+    }
+
+    float sum = 0.0f;
+    for (size_t i = 0; i < v1.size(); ++i) {
+        float diff = v1[i] - v2[i];
+        sum += diff * diff;
+    }
+    return std::sqrt(sum);
+}
+
+float RAGSimilaritySearch::cosineSimilarity(const RAGVector &v1, const RAGVector &v2)
+{
+    if (v1.size() != v2.size()) {
+        LOG_MESSAGE(QString("Vector size mismatch: %1 vs %2").arg(v1.size()).arg(v2.size()));
+        return 0.0f;
+    }
+
+    float dotProduct = 0.0f;
+    float norm1 = 0.0f;
+    float norm2 = 0.0f;
+
+    for (size_t i = 0; i < v1.size(); ++i) {
+        dotProduct += v1[i] * v2[i];
+        norm1 += v1[i] * v1[i];
+        norm2 += v2[i] * v2[i];
+    }
+
+    norm1 = std::sqrt(norm1);
+    norm2 = std::sqrt(norm2);
+
+    if (norm1 == 0.0f || norm2 == 0.0f)
+        return 0.0f;
+    return dotProduct / (norm1 * norm2);
+}
+
+} // namespace QodeAssist::Context

context/RAGSimilaritySearch.hpp

@@ -17,17 +17,21 @@
  * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
  */
 
-namespace QodeAssist::LLMCore {
+#pragma once
 
-enum class ProviderID {
-    Any,
-    Ollama,
-    LMStudio,
-    Claude,
-    OpenAI,
-    OpenAICompatible,
-    MistralAI,
-    OpenRouter,
-    GoogleAI
+#include "RAGData.hpp"
+
+namespace QodeAssist::Context {
+
+class RAGSimilaritySearch
+{
+public:
+    static float l2Distance(const RAGVector &v1, const RAGVector &v2);
+
+    static float cosineSimilarity(const RAGVector &v1, const RAGVector &v2);
+
+private:
+    RAGSimilaritySearch() = delete;
 };
-}
+
+} // namespace QodeAssist::Context

context/RAGStorage.hpp

@@ -0,0 +1,174 @@
+/*
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+// RAGStorage.hpp
+#pragma once
+
+#include <optional>
+#include <QDateTime>
+#include <QMutex>
+#include <QObject>
+#include <QSqlDatabase>
+#include <QString>
+#include <qsqlquery.h>
+
+#include <RAGData.hpp>
+
+namespace QodeAssist::Context {
+
+struct FileChunkData
+{
+    QString filePath;
+    int startLine;
+    int endLine;
+    QString content;
+    QDateTime createdAt;
+    QDateTime updatedAt;
+
+    bool isValid() const
+    {
+        return !filePath.isEmpty() && startLine >= 0 && endLine >= startLine && !content.isEmpty();
+    }
+};
+
+struct StorageOptions
+{
+    int maxChunkSize = 1024 * 1024;
+    int maxVectorSize = 1024;
+    bool useCompression = false;
+    bool enableLogging = false;
+};
+
+struct StorageStatistics
+{
+    int totalChunks;
+    int totalVectors;
+    int totalFiles;
+    qint64 totalSize;
+    QDateTime lastUpdate;
+};
+
+class RAGStorage : public QObject
+{
+    Q_OBJECT
+
+public:
+    static constexpr int CURRENT_VERSION = 1;
+
+    enum class Status { Ok, DatabaseError, ValidationError, VersionError, ConnectionError };
+
+    struct ValidationResult
+    {
+        bool isValid;
+        QString errorMessage;
+        Status errorStatus;
+    };
+
+    struct Error
+    {
+        QString message;
+        QString sqlError;
+        QString query;
+        Status status;
+    };
+
+    explicit RAGStorage(
+        const QString &dbPath,
+        const StorageOptions &options = StorageOptions(),
+        QObject *parent = nullptr);
+    ~RAGStorage();
+
+    bool init();
+    Status status() const;
+    Error lastError() const;
+    bool isReady() const;
+    QString dbPath() const;
+
+    bool beginTransaction();
+    bool commitTransaction();
+    bool rollbackTransaction();
+
+    bool storeVector(const QString &filePath, const RAGVector &vector);
+    bool updateVector(const QString &filePath, const RAGVector &vector);
+    std::optional<RAGVector> getVector(const QString &filePath);
+    bool needsUpdate(const QString &filePath);
+    QStringList getAllFiles();
+
+    bool storeChunk(const FileChunkData &chunk);
+    bool storeChunks(const QList<FileChunkData> &chunks);
+    bool updateChunk(const FileChunkData &chunk);
+    bool updateChunks(const QList<FileChunkData> &chunks);
+    bool deleteChunksForFile(const QString &filePath);
+    std::optional<FileChunkData> getChunk(const QString &filePath, int startLine, int endLine);
+    QList<FileChunkData> getChunksForFile(const QString &filePath);
+    bool chunkExists(const QString &filePath, int startLine, int endLine);
+
+    int getChunkCount(const QString &filePath);
+    bool deleteOldChunks(const QString &filePath, const QDateTime &olderThan);
+    bool deleteAllChunks();
+    QStringList getFilesWithChunks();
+    bool vacuum();
+    bool backup(const QString &backupPath);
+    bool restore(const QString &backupPath);
+    StorageStatistics getStatistics() const;
+
+    int getStorageVersion() const;
+    bool isVersionCompatible() const;
+
+    bool applyMigration(int version);
+signals:
+    void errorOccurred(const Error &error);
+    void operationCompleted(const QString &operation);
+
+private:
+    bool createTables();
+    bool createIndices();
+    bool createVersionTable();
+    bool createChunksTable();
+    bool createVectorsTable();
+    bool openDatabase();
+    bool initializeNewStorage();
+    bool upgradeStorage(int fromVersion);
+    bool validateSchema() const;
+
+    QDateTime getFileLastModified(const QString &filePath);
+    RAGVector blobToVector(const QByteArray &blob);
+    QByteArray vectorToBlob(const RAGVector &vector);
+
+    void setError(const QString &message, Status status = Status::DatabaseError);
+    void clearError();
+    bool prepareStatements();
+    ValidationResult validateChunk(const FileChunkData &chunk) const;
+    ValidationResult validateVector(const RAGVector &vector) const;
+
+private:
+    QSqlDatabase m_db;
+    QString m_dbPath;
+    StorageOptions m_options;
+    mutable QMutex m_mutex;
+    Error m_lastError;
+    Status m_status;
+
+    QSqlQuery m_insertChunkQuery;
+    QSqlQuery m_updateChunkQuery;
+    QSqlQuery m_insertVectorQuery;
+    QSqlQuery m_updateVectorQuery;
+};
+
+} // namespace QodeAssist::Context

context/RAGVectorizer.cpp

@@ -0,0 +1,116 @@
+/*
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "RAGVectorizer.hpp"
+
+#include <QJsonArray>
+#include <QJsonDocument>
+#include <QJsonObject>
+#include <QNetworkReply>
+
+namespace QodeAssist::Context {
+
+RAGVectorizer::RAGVectorizer(const QString &providerUrl,
+                             const QString &modelName,
+                             QObject *parent)
+    : QObject(parent)
+    , m_network(new QNetworkAccessManager(this))
+    , m_embedProviderUrl(providerUrl)
+    , m_model(modelName)
+{}
+
+RAGVectorizer::~RAGVectorizer() {}
+
+QJsonObject RAGVectorizer::prepareEmbeddingRequest(const QString &text) const
+{
+    return QJsonObject{{"model", m_model}, {"prompt", text}};
+}
+
+RAGVector RAGVectorizer::parseEmbeddingResponse(const QByteArray &response) const
+{
+    QJsonDocument doc = QJsonDocument::fromJson(response);
+    if (doc.isNull()) {
+        qDebug() << "Failed to parse JSON response";
+        return RAGVector();
+    }
+
+    QJsonObject obj = doc.object();
+    if (!obj.contains("embedding")) {
+        qDebug() << "Response does not contain 'embedding' field";
+        // qDebug() << "Response content:" << response;
+        return RAGVector();
+    }
+
+    QJsonArray array = obj["embedding"].toArray();
+    if (array.isEmpty()) {
+        qDebug() << "Embedding array is empty";
+        return RAGVector();
+    }
+
+    RAGVector result;
+    result.reserve(array.size());
+    for (const auto &value : array) {
+        result.push_back(value.toDouble());
+    }
+
+    qDebug() << "Successfully parsed vector with size:" << result.size();
+    return result;
+}
+
+QFuture<RAGVector> RAGVectorizer::vectorizeText(const QString &text)
+{
+    qDebug() << "Vectorizing text, length:" << text.length();
+    qDebug() << "Using embedding provider:" << m_embedProviderUrl;
+
+    auto promise = std::make_shared<QPromise<RAGVector>>();
+    promise->start();
+
+    QNetworkRequest request(QUrl(m_embedProviderUrl + "/api/embeddings"));
+    request.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
+
+    QJsonObject requestData = prepareEmbeddingRequest(text);
+    QByteArray jsonData = QJsonDocument(requestData).toJson();
+    qDebug() << "Sending request to embeddings API:" << jsonData;
+
+    auto reply = m_network->post(request, jsonData);
+
+    connect(reply, &QNetworkReply::finished, this, [promise, reply, this]() {
+        if (reply->error() == QNetworkReply::NoError) {
+            QByteArray response = reply->readAll();
+            // qDebug() << "Received response from embeddings API:" << response;
+
+            auto vector = parseEmbeddingResponse(response);
+            qDebug() << "Parsed vector size:" << vector.size();
+            promise->addResult(vector);
+        } else {
+            qDebug() << "Network error:" << reply->errorString();
+            qDebug() << "HTTP status code:"
+                     << reply->attribute(QNetworkRequest::HttpStatusCodeAttribute).toInt();
+            qDebug() << "Response:" << reply->readAll();
+
+            promise->addResult(RAGVector());
+        }
+        promise->finish();
+        reply->deleteLater();
+    });
+
+    return promise->future();
+}
+
+} // namespace QodeAssist::Context

context/RAGVectorizer.hpp

@@ -0,0 +1,51 @@
+/*
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <QFuture>
+#include <QNetworkAccessManager>
+#include <QObject>
+
+#include <RAGData.hpp>
+
+namespace QodeAssist::Context {
+
+class RAGVectorizer : public QObject
+{
+    Q_OBJECT
+public:
+    explicit RAGVectorizer(
+        const QString &providerUrl = "http://localhost:11434",
+        const QString &modelName = "all-minilm:33m-l12-v2-fp16",
+        QObject *parent = nullptr);
+    ~RAGVectorizer();
+
+    QFuture<RAGVector> vectorizeText(const QString &text);
+
+private:
+    QJsonObject prepareEmbeddingRequest(const QString &text) const;
+    RAGVector parseEmbeddingResponse(const QByteArray &response) const;
+
+    QNetworkAccessManager *m_network;
+    QString m_embedProviderUrl;
+    QString m_model;
+};
+
+} // namespace QodeAssist::Context

llmcore/CMakeLists.txt

@@ -10,7 +10,7 @@ add_library(LLMCore STATIC
     OllamaMessage.hpp OllamaMessage.cpp
     OpenAIMessage.hpp OpenAIMessage.cpp
     ValidationUtils.hpp ValidationUtils.cpp
-    ProviderID.hpp
+    MessageBuilder.hpp MessageBuilder.cpp
 )
 
 target_link_libraries(LLMCore

llmcore/ContextData.hpp

@@ -20,23 +20,14 @@
 #pragma once
 
 #include <QString>
-#include <QVector>
 
 namespace QodeAssist::LLMCore {
 
-struct Message
-{
-    QString role;
-    QString content;
-};
-
 struct ContextData
 {
-    std::optional<QString> systemPrompt;
-    std::optional<QString> prefix;
-    std::optional<QString> suffix;
-    std::optional<QString> fileContext;
-    std::optional<QVector<Message>> history;
+    QString prefix;
+    QString suffix;
+    QString fileContext;
 };
 
 } // namespace QodeAssist::LLMCore

llmcore/MessageBuilder.cpp

@@ -0,0 +1,93 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "MessageBuilder.hpp"
+
+QodeAssist::LLMCore::MessageBuilder &QodeAssist::LLMCore::MessageBuilder::addSystemMessage(
+    const QString &content)
+{
+    m_systemMessage = content;
+    return *this;
+}
+
+QodeAssist::LLMCore::MessageBuilder &QodeAssist::LLMCore::MessageBuilder::addUserMessage(
+    const QString &content)
+{
+    m_messages.append({MessageRole::User, content});
+    return *this;
+}
+
+QodeAssist::LLMCore::MessageBuilder &QodeAssist::LLMCore::MessageBuilder::addSuffix(
+    const QString &content)
+{
+    m_suffix = content;
+    return *this;
+}
+
+QodeAssist::LLMCore::MessageBuilder &QodeAssist::LLMCore::MessageBuilder::addTokenizer(
+    PromptTemplate *promptTemplate)
+{
+    m_promptTemplate = promptTemplate;
+    return *this;
+}
+
+QString QodeAssist::LLMCore::MessageBuilder::roleToString(MessageRole role) const
+{
+    switch (role) {
+    case MessageRole::System:
+        return ROLE_SYSTEM;
+    case MessageRole::User:
+        return ROLE_USER;
+    case MessageRole::Assistant:
+        return ROLE_ASSISTANT;
+    default:
+        return ROLE_USER;
+    }
+}
+
+void QodeAssist::LLMCore::MessageBuilder::saveTo(QJsonObject &request, ProvidersApi api)
+{
+    if (!m_promptTemplate) {
+        return;
+    }
+
+    ContextData context{
+        m_messages.isEmpty() ? QString() : m_messages.last().content, m_suffix, m_systemMessage};
+
+    if (api == ProvidersApi::Ollama) {
+        if (m_promptTemplate->type() == TemplateType::Fim) {
+            request["system"] = m_systemMessage;
+            m_promptTemplate->prepareRequest(request, context);
+        } else {
+            QJsonArray messages;
+
+            messages.append(QJsonObject{{"role", "system"}, {"content", m_systemMessage}});
+            messages.append(QJsonObject{{"role", "user"}, {"content", m_messages.last().content}});
+            request["messages"] = messages;
+            m_promptTemplate->prepareRequest(request, context);
+        }
+    } else if (api == ProvidersApi::OpenAI) {
+        QJsonArray messages;
+
+        messages.append(QJsonObject{{"role", "system"}, {"content", m_systemMessage}});
+        messages.append(QJsonObject{{"role", "user"}, {"content", m_messages.last().content}});
+        request["messages"] = messages;
+        m_promptTemplate->prepareRequest(request, context);
+    }
+}

llmcore/MessageBuilder.hpp

@@ -0,0 +1,68 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <QJsonArray>
+#include <QJsonObject>
+#include <QString>
+#include <QVector>
+
+#include "PromptTemplate.hpp"
+
+namespace QodeAssist::LLMCore {
+
+enum class MessageRole { System, User, Assistant };
+
+enum class OllamaFormat { Messages, Completions };
+
+enum class ProvidersApi { Ollama, OpenAI, Claude };
+
+static const QString ROLE_SYSTEM = "system";
+static const QString ROLE_USER = "user";
+static const QString ROLE_ASSISTANT = "assistant";
+
+struct Message
+{
+    MessageRole role;
+    QString content;
+};
+
+class MessageBuilder
+{
+public:
+    MessageBuilder &addSystemMessage(const QString &content);
+
+    MessageBuilder &addUserMessage(const QString &content);
+
+    MessageBuilder &addSuffix(const QString &content);
+
+    MessageBuilder &addTokenizer(PromptTemplate *promptTemplate);
+
+    QString roleToString(MessageRole role) const;
+
+    void saveTo(QJsonObject &request, ProvidersApi api);
+
+private:
+    QString m_systemMessage;
+    QString m_suffix;
+    QVector<Message> m_messages;
+    PromptTemplate *m_promptTemplate;
+};
+} // namespace QodeAssist::LLMCore

llmcore/PromptTemplate.hpp

@@ -24,11 +24,10 @@
 #include <QString>
 
 #include "ContextData.hpp"
-#include "ProviderID.hpp"
 
 namespace QodeAssist::LLMCore {
 
-enum class TemplateType { Chat, FIM };
+enum class TemplateType { Chat, Fim };
 
 class PromptTemplate
 {
@@ -36,9 +35,9 @@ public:
     virtual ~PromptTemplate() = default;
     virtual TemplateType type() const = 0;
     virtual QString name() const = 0;
+    virtual QString promptTemplate() const = 0;
     virtual QStringList stopWords() const = 0;
     virtual void prepareRequest(QJsonObject &request, const ContextData &context) const = 0;
     virtual QString description() const = 0;
-    virtual bool isSupportProvider(ProviderID id) const = 0;
 };
 } // namespace QodeAssist::LLMCore

llmcore/PromptTemplateManager.cpp

@@ -37,32 +37,6 @@ QStringList PromptTemplateManager::chatTemplatesNames() const
     return m_chatTemplates.keys();
 }
 
-QStringList PromptTemplateManager::getFimTemplatesForProvider(ProviderID id)
-{
-    QStringList templateList;
-
-    for (const auto tmpl : m_fimTemplates) {
-        if (tmpl->isSupportProvider(id)) {
-            templateList.append(tmpl->name());
-        }
-    }
-
-    return templateList;
-}
-
-QStringList PromptTemplateManager::getChatTemplatesForProvider(ProviderID id)
-{
-    QStringList templateList;
-
-    for (const auto tmpl : m_chatTemplates) {
-        if (tmpl->isSupportProvider(id)) {
-            templateList.append(tmpl->name());
-        }
-    }
-
-    return templateList;
-}
-
 PromptTemplateManager::~PromptTemplateManager()
 {
     qDeleteAll(m_fimTemplates);
@@ -70,15 +44,11 @@ PromptTemplateManager::~PromptTemplateManager()
 
 PromptTemplate *PromptTemplateManager::getFimTemplateByName(const QString &templateName)
 {
-    if (!m_fimTemplates.contains(templateName))
-        return m_fimTemplates.first();
     return m_fimTemplates[templateName];
 }
 
 PromptTemplate *PromptTemplateManager::getChatTemplateByName(const QString &templateName)
 {
-    if (!m_chatTemplates.contains(templateName))
-        return m_chatTemplates.first();
     return m_chatTemplates[templateName];
 }
 

llmcore/PromptTemplateManager.hpp

@@ -51,9 +51,6 @@ public:
     QStringList fimTemplatesNames() const;
     QStringList chatTemplatesNames() const;
 
-    QStringList getFimTemplatesForProvider(ProviderID id);
-    QStringList getChatTemplatesForProvider(ProviderID id);
-
 private:
     PromptTemplateManager() = default;
     PromptTemplateManager(const PromptTemplateManager &) = delete;

llmcore/Provider.hpp

@@ -23,7 +23,6 @@
 #include <QNetworkRequest>
 #include <QString>
 
-#include "ContextData.hpp"
 #include "PromptTemplate.hpp"
 #include "RequestType.hpp"
 
@@ -42,18 +41,13 @@ public:
     virtual QString completionEndpoint() const = 0;
     virtual QString chatEndpoint() const = 0;
     virtual bool supportsModelListing() const = 0;
-    virtual void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type)
-        = 0;
+
+    virtual void prepareRequest(QJsonObject &request, RequestType type) = 0;
     virtual bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) = 0;
     virtual QList<QString> getInstalledModels(const QString &url) = 0;
     virtual QList<QString> validateRequest(const QJsonObject &request, TemplateType type) = 0;
     virtual QString apiKey() const = 0;
     virtual void prepareNetworkRequest(QNetworkRequest &networkRequest) const = 0;
-    virtual ProviderID providerID() const = 0;
 };
 
 } // namespace QodeAssist::LLMCore

llmcore/ProvidersManager.cpp

@@ -39,8 +39,6 @@ ProvidersManager::~ProvidersManager()
 
 Provider *ProvidersManager::getProviderByName(const QString &providerName)
 {
-    if (!m_providers.contains(providerName))
-        return m_providers.first();
     return m_providers[providerName];
 }
 

llmcore/RequestHandler.cpp

@@ -58,10 +58,7 @@ void RequestHandler::sendLLMRequest(const LLMConfig &config, const QJsonObject &
         reply->deleteLater();
         m_activeRequests.remove(requestId);
         if (reply->error() != QNetworkReply::NoError) {
-            LOG_MESSAGE(QString("Error details: %1\nStatus code: %2\nResponse: %3")
-                            .arg(reply->errorString())
-                            .arg(reply->attribute(QNetworkRequest::HttpStatusCodeAttribute).toInt())
-                            .arg(QString(reply->readAll())));
+            LOG_MESSAGE(QString("Error in QodeAssist request: %1").arg(reply->errorString()));
             emit requestFinished(requestId, false, reply->errorString());
         } else {
             LOG_MESSAGE("Request finished successfully");

llmcore/RequestType.hpp

@@ -21,5 +21,5 @@
 
 namespace QodeAssist::LLMCore {
 
-enum RequestType { CodeCompletion, Chat, Embedding };
+enum RequestType { CodeCompletion, Chat };
 }

providers/ClaudeProvider.cpp

@@ -30,10 +30,13 @@
 #include "logger/Logger.hpp"
 #include "settings/ChatAssistantSettings.hpp"
 #include "settings/CodeCompletionSettings.hpp"
+#include "settings/GeneralSettings.hpp"
 #include "settings/ProviderSettings.hpp"
 
 namespace QodeAssist::Providers {
 
+ClaudeProvider::ClaudeProvider() {}
+
 QString ClaudeProvider::name() const
 {
     return "Claude";
@@ -59,17 +62,31 @@ bool ClaudeProvider::supportsModelListing() const
     return true;
 }
 
-void ClaudeProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
+void ClaudeProvider::prepareRequest(QJsonObject &request, LLMCore::RequestType type)
 {
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
+    auto prepareMessages = [](QJsonObject &req) -> QJsonArray {
+        QJsonArray messages;
+        if (req.contains("messages")) {
+            QJsonArray origMessages = req["messages"].toArray();
+            for (const auto &msg : origMessages) {
+                QJsonObject message = msg.toObject();
+                if (message["role"].toString() == "system") {
+                    req["system"] = message["content"];
+                } else {
+                    messages.append(message);
+                }
+            }
+        } else {
+            if (req.contains("system")) {
+                req["system"] = req["system"].toString();
+            }
+            if (req.contains("prompt")) {
+                messages.append(
+                    QJsonObject{{"role", "user"}, {"content", req.take("prompt").toString()}});
+            }
+        }
+        return messages;
+    };
 
     auto applyModelParams = [&request](const auto &settings) {
         request["max_tokens"] = settings.maxTokens();
@@ -81,6 +98,11 @@ void ClaudeProvider::prepareRequest(
         request["stream"] = true;
     };
 
+    QJsonArray messages = prepareMessages(request);
+    if (!messages.isEmpty()) {
+        request["messages"] = std::move(messages);
+    }
+
     if (type == LLMCore::RequestType::CodeCompletion) {
         applyModelParams(Settings::codeCompletionSettings());
     } else {
@@ -213,9 +235,4 @@ void ClaudeProvider::prepareNetworkRequest(QNetworkRequest &networkRequest) cons
     }
 }
 
-LLMCore::ProviderID ClaudeProvider::providerID() const
-{
-    return LLMCore::ProviderID::Claude;
-}
-
 } // namespace QodeAssist::Providers

providers/ClaudeProvider.hpp

@@ -26,22 +26,19 @@ namespace QodeAssist::Providers {
 class ClaudeProvider : public LLMCore::Provider
 {
 public:
+    ClaudeProvider();
+
     QString name() const override;
     QString url() const override;
     QString completionEndpoint() const override;
     QString chatEndpoint() const override;
     bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
+    void prepareRequest(QJsonObject &request, LLMCore::RequestType type) override;
     bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
     QList<QString> getInstalledModels(const QString &url) override;
     QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
     QString apiKey() const override;
     void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
 };
 
 } // namespace QodeAssist::Providers

providers/GoogleAIProvider.cpp

@@ -1,303 +0,0 @@
-/* 
- * Copyright (C) 2024 Petr Mironychev
- *
- * This file is part of QodeAssist.
- *
- * QodeAssist is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * QodeAssist is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#include "GoogleAIProvider.hpp"
-
-#include <QEventLoop>
-#include <QJsonArray>
-#include <QJsonDocument>
-#include <QJsonObject>
-#include <QNetworkReply>
-#include <QtCore/qurlquery.h>
-
-#include "llmcore/ValidationUtils.hpp"
-#include "logger/Logger.hpp"
-#include "settings/ChatAssistantSettings.hpp"
-#include "settings/CodeCompletionSettings.hpp"
-#include "settings/GeneralSettings.hpp"
-#include "settings/ProviderSettings.hpp"
-
-namespace QodeAssist::Providers {
-
-QString GoogleAIProvider::name() const
-{
-    return "Google AI";
-}
-
-QString GoogleAIProvider::url() const
-{
-    return "https://generativelanguage.googleapis.com/v1beta";
-}
-
-QString GoogleAIProvider::completionEndpoint() const
-{
-    return {};
-}
-
-QString GoogleAIProvider::chatEndpoint() const
-{
-    return {};
-}
-
-bool GoogleAIProvider::supportsModelListing() const
-{
-    return true;
-}
-
-void GoogleAIProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
-{
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
-
-    auto applyModelParams = [&request](const auto &settings) {
-        QJsonObject generationConfig;
-        generationConfig["maxOutputTokens"] = settings.maxTokens();
-        generationConfig["temperature"] = settings.temperature();
-
-        if (settings.useTopP())
-            generationConfig["topP"] = settings.topP();
-        if (settings.useTopK())
-            generationConfig["topK"] = settings.topK();
-
-        request["generationConfig"] = generationConfig;
-    };
-
-    if (type == LLMCore::RequestType::CodeCompletion) {
-        applyModelParams(Settings::codeCompletionSettings());
-    } else {
-        applyModelParams(Settings::chatAssistantSettings());
-    }
-}
-
-bool GoogleAIProvider::handleResponse(QNetworkReply *reply, QString &accumulatedResponse)
-{
-    if (reply->isFinished()) {
-        if (reply->bytesAvailable() > 0) {
-            QByteArray data = reply->readAll();
-
-            if (data.startsWith("data: ")) {
-                return handleStreamResponse(data, accumulatedResponse);
-            } else {
-                return handleRegularResponse(data, accumulatedResponse);
-            }
-        }
-
-        return true;
-    }
-
-    QByteArray data = reply->readAll();
-    if (data.isEmpty()) {
-        return false;
-    }
-
-    if (data.startsWith("data: ")) {
-        return handleStreamResponse(data, accumulatedResponse);
-    } else {
-        return handleRegularResponse(data, accumulatedResponse);
-    }
-}
-
-QList<QString> GoogleAIProvider::getInstalledModels(const QString &url)
-{
-    QList<QString> models;
-
-    QNetworkAccessManager manager;
-    QNetworkRequest request(QString("%1/models?key=%2").arg(url, apiKey()));
-
-    request.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
-
-    QNetworkReply *reply = manager.get(request);
-    QEventLoop loop;
-    QObject::connect(reply, &QNetworkReply::finished, &loop, &QEventLoop::quit);
-    loop.exec();
-
-    if (reply->error() == QNetworkReply::NoError) {
-        QByteArray responseData = reply->readAll();
-        QJsonDocument jsonResponse = QJsonDocument::fromJson(responseData);
-        QJsonObject jsonObject = jsonResponse.object();
-
-        if (jsonObject.contains("models")) {
-            QJsonArray modelArray = jsonObject["models"].toArray();
-            models.clear();
-
-            for (const QJsonValue &value : modelArray) {
-                QJsonObject modelObject = value.toObject();
-                if (modelObject.contains("name")) {
-                    QString modelName = modelObject["name"].toString();
-                    if (modelName.contains("/")) {
-                        modelName = modelName.split("/").last();
-                    }
-                    models.append(modelName);
-                }
-            }
-        }
-    } else {
-        LOG_MESSAGE(QString("Error fetching Google AI models: %1").arg(reply->errorString()));
-    }
-
-    reply->deleteLater();
-    return models;
-}
-
-QList<QString> GoogleAIProvider::validateRequest(
-    const QJsonObject &request, LLMCore::TemplateType type)
-{
-    QJsonObject templateReq;
-
-    templateReq = QJsonObject{
-        {"contents", QJsonArray{}},
-        {"system_instruction", QJsonArray{}},
-        {"generationConfig",
-         QJsonObject{{"temperature", {}}, {"maxOutputTokens", {}}, {"topP", {}}, {"topK", {}}}},
-        {"safetySettings", QJsonArray{}}};
-
-    return LLMCore::ValidationUtils::validateRequestFields(request, templateReq);
-}
-
-QString GoogleAIProvider::apiKey() const
-{
-    return Settings::providerSettings().googleAiApiKey();
-}
-
-void GoogleAIProvider::prepareNetworkRequest(QNetworkRequest &networkRequest) const
-{
-    networkRequest.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
-
-    QUrl url = networkRequest.url();
-    QUrlQuery query(url.query());
-    query.addQueryItem("key", apiKey());
-    url.setQuery(query);
-    networkRequest.setUrl(url);
-}
-
-LLMCore::ProviderID GoogleAIProvider::providerID() const
-{
-    return LLMCore::ProviderID::GoogleAI;
-}
-
-bool GoogleAIProvider::handleStreamResponse(const QByteArray &data, QString &accumulatedResponse)
-{
-    QByteArrayList lines = data.split('\n');
-    bool isDone = false;
-
-    for (const QByteArray &line : lines) {
-        QByteArray trimmedLine = line.trimmed();
-        if (trimmedLine.isEmpty()) {
-            continue;
-        }
-
-        if (trimmedLine == "data: [DONE]") {
-            isDone = true;
-            continue;
-        }
-
-        if (trimmedLine.startsWith("data: ")) {
-            QByteArray jsonData = trimmedLine.mid(6); // Remove "data: " prefix
-            QJsonDocument doc = QJsonDocument::fromJson(jsonData);
-            if (doc.isNull() || !doc.isObject()) {
-                continue;
-            }
-
-            QJsonObject responseObj = doc.object();
-
-            if (responseObj.contains("error")) {
-                QJsonObject error = responseObj["error"].toObject();
-                LOG_MESSAGE("Error in Google AI stream response: " + error["message"].toString());
-                continue;
-            }
-
-            if (responseObj.contains("candidates")) {
-                QJsonArray candidates = responseObj["candidates"].toArray();
-                if (!candidates.isEmpty()) {
-                    QJsonObject candidate = candidates.first().toObject();
-
-                    if (candidate.contains("finishReason")
-                        && !candidate["finishReason"].toString().isEmpty()) {
-                        isDone = true;
-                    }
-
-                    if (candidate.contains("content")) {
-                        QJsonObject content = candidate["content"].toObject();
-                        if (content.contains("parts")) {
-                            QJsonArray parts = content["parts"].toArray();
-                            for (const auto &part : parts) {
-                                QJsonObject partObj = part.toObject();
-                                if (partObj.contains("text")) {
-                                    accumulatedResponse += partObj["text"].toString();
-                                }
-                            }
-                        }
-                    }
-                }
-            }
-        }
-    }
-
-    return isDone;
-}
-
-bool GoogleAIProvider::handleRegularResponse(const QByteArray &data, QString &accumulatedResponse)
-{
-    QJsonDocument doc = QJsonDocument::fromJson(data);
-    if (doc.isNull() || !doc.isObject()) {
-        LOG_MESSAGE("Invalid JSON response from Google AI API");
-        return false;
-    }
-
-    QJsonObject response = doc.object();
-
-    if (response.contains("error")) {
-        QJsonObject error = response["error"].toObject();
-        LOG_MESSAGE("Error in Google AI response: " + error["message"].toString());
-        return false;
-    }
-
-    if (!response.contains("candidates") || response["candidates"].toArray().isEmpty()) {
-        return false;
-    }
-
-    QJsonObject candidate = response["candidates"].toArray().first().toObject();
-    if (!candidate.contains("content")) {
-        return false;
-    }
-
-    QJsonObject content = candidate["content"].toObject();
-    if (!content.contains("parts")) {
-        return false;
-    }
-
-    QJsonArray parts = content["parts"].toArray();
-    for (const auto &part : parts) {
-        QJsonObject partObj = part.toObject();
-        if (partObj.contains("text")) {
-            accumulatedResponse += partObj["text"].toString();
-        }
-    }
-
-    return true;
-}
-
-} // namespace QodeAssist::Providers

providers/GoogleAIProvider.hpp

@@ -1,51 +0,0 @@
-/* 
- * Copyright (C) 2024 Petr Mironychev
- *
- * This file is part of QodeAssist.
- *
- * QodeAssist is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * QodeAssist is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#pragma once
-
-#include "llmcore/Provider.hpp"
-
-namespace QodeAssist::Providers {
-
-class GoogleAIProvider : public LLMCore::Provider
-{
-public:
-    QString name() const override;
-    QString url() const override;
-    QString completionEndpoint() const override;
-    QString chatEndpoint() const override;
-    bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
-    bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
-    QList<QString> getInstalledModels(const QString &url) override;
-    QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
-    QString apiKey() const override;
-    void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
-
-private:
-    bool handleStreamResponse(const QByteArray &data, QString &accumulatedResponse);
-    bool handleRegularResponse(const QByteArray &data, QString &accumulatedResponse);
-};
-
-} // namespace QodeAssist::Providers

providers/LMStudioProvider.cpp

@@ -33,6 +33,8 @@
 
 namespace QodeAssist::Providers {
 
+LMStudioProvider::LMStudioProvider() {}
+
 QString LMStudioProvider::name() const
 {
     return "LM Studio";
@@ -58,6 +60,47 @@ bool LMStudioProvider::supportsModelListing() const
     return true;
 }
 
+void LMStudioProvider::prepareRequest(QJsonObject &request, LLMCore::RequestType type)
+{
+    auto prepareMessages = [](QJsonObject &req) -> QJsonArray {
+        QJsonArray messages;
+        if (req.contains("system")) {
+            messages.append(
+                QJsonObject{{"role", "system"}, {"content", req.take("system").toString()}});
+        }
+        if (req.contains("prompt")) {
+            messages.append(
+                QJsonObject{{"role", "user"}, {"content", req.take("prompt").toString()}});
+        }
+        return messages;
+    };
+
+    auto applyModelParams = [&request](const auto &settings) {
+        request["max_tokens"] = settings.maxTokens();
+        request["temperature"] = settings.temperature();
+
+        if (settings.useTopP())
+            request["top_p"] = settings.topP();
+        if (settings.useTopK())
+            request["top_k"] = settings.topK();
+        if (settings.useFrequencyPenalty())
+            request["frequency_penalty"] = settings.frequencyPenalty();
+        if (settings.usePresencePenalty())
+            request["presence_penalty"] = settings.presencePenalty();
+    };
+
+    QJsonArray messages = prepareMessages(request);
+    if (!messages.isEmpty()) {
+        request["messages"] = std::move(messages);
+    }
+
+    if (type == LLMCore::RequestType::CodeCompletion) {
+        applyModelParams(Settings::codeCompletionSettings());
+    } else {
+        applyModelParams(Settings::chatAssistantSettings());
+    }
+}
+
 bool LMStudioProvider::handleResponse(QNetworkReply *reply, QString &accumulatedResponse)
 {
     QByteArray data = reply->readAll();
@@ -168,42 +211,4 @@ void LMStudioProvider::prepareNetworkRequest(QNetworkRequest &networkRequest) co
     networkRequest.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
 }
 
-LLMCore::ProviderID LMStudioProvider::providerID() const
-{
-    return LLMCore::ProviderID::LMStudio;
-}
-
-void QodeAssist::Providers::LMStudioProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
-{
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
-
-    auto applyModelParams = [&request](const auto &settings) {
-        request["max_tokens"] = settings.maxTokens();
-        request["temperature"] = settings.temperature();
-
-        if (settings.useTopP())
-            request["top_p"] = settings.topP();
-        if (settings.useTopK())
-            request["top_k"] = settings.topK();
-        if (settings.useFrequencyPenalty())
-            request["frequency_penalty"] = settings.frequencyPenalty();
-        if (settings.usePresencePenalty())
-            request["presence_penalty"] = settings.presencePenalty();
-    };
-
-    if (type == LLMCore::RequestType::CodeCompletion) {
-        applyModelParams(Settings::codeCompletionSettings());
-    } else {
-        applyModelParams(Settings::chatAssistantSettings());
-    }
-}
-
 } // namespace QodeAssist::Providers

providers/LMStudioProvider.hpp

@@ -26,22 +26,19 @@ namespace QodeAssist::Providers {
 class LMStudioProvider : public LLMCore::Provider
 {
 public:
+    LMStudioProvider();
+
     QString name() const override;
     QString url() const override;
     QString completionEndpoint() const override;
     QString chatEndpoint() const override;
     bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
+    void prepareRequest(QJsonObject &request, LLMCore::RequestType type) override;
     bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
     QList<QString> getInstalledModels(const QString &url) override;
     QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
     QString apiKey() const override;
     void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
 };
 
 } // namespace QodeAssist::Providers

providers/MistralAIProvider.cpp

@@ -1,220 +0,0 @@
-#include "MistralAIProvider.hpp"
-
-#include "settings/ChatAssistantSettings.hpp"
-#include "settings/CodeCompletionSettings.hpp"
-#include "settings/ProviderSettings.hpp"
-
-#include <QJsonArray>
-#include <QJsonDocument>
-#include <QJsonObject>
-#include <QNetworkReply>
-#include <QtCore/qeventloop.h>
-
-#include "llmcore/OpenAIMessage.hpp"
-#include "llmcore/ValidationUtils.hpp"
-#include "logger/Logger.hpp"
-
-namespace QodeAssist::Providers {
-
-QString MistralAIProvider::name() const
-{
-    return "Mistral AI";
-}
-
-QString MistralAIProvider::url() const
-{
-    return "https://api.mistral.ai";
-}
-
-QString MistralAIProvider::completionEndpoint() const
-{
-    return "/v1/fim/completions";
-}
-
-QString MistralAIProvider::chatEndpoint() const
-{
-    return "/v1/chat/completions";
-}
-
-bool MistralAIProvider::supportsModelListing() const
-{
-    return true;
-}
-
-bool MistralAIProvider::handleResponse(QNetworkReply *reply, QString &accumulatedResponse)
-{
-    QByteArray data = reply->readAll();
-    if (data.isEmpty()) {
-        return false;
-    }
-
-    bool isDone = false;
-    QByteArrayList lines = data.split('\n');
-
-    for (const QByteArray &line : lines) {
-        if (line.trimmed().isEmpty()) {
-            continue;
-        }
-
-        if (line == "data: [DONE]") {
-            isDone = true;
-            continue;
-        }
-
-        QByteArray jsonData = line;
-        if (line.startsWith("data: ")) {
-            jsonData = line.mid(6);
-        }
-
-        QJsonParseError error;
-        QJsonDocument doc = QJsonDocument::fromJson(jsonData, &error);
-
-        if (doc.isNull()) {
-            continue;
-        }
-
-        auto message = LLMCore::OpenAIMessage::fromJson(doc.object());
-        if (message.hasError()) {
-            LOG_MESSAGE("Error in OpenAI response: " + message.error);
-            continue;
-        }
-
-        QString content = message.getContent();
-        if (!content.isEmpty()) {
-            accumulatedResponse += content;
-        }
-
-        if (message.isDone()) {
-            isDone = true;
-        }
-    }
-
-    return isDone;
-}
-
-QList<QString> MistralAIProvider::getInstalledModels(const QString &url)
-{
-    QList<QString> models;
-    QNetworkAccessManager manager;
-    QNetworkRequest request(QString("%1/v1/models").arg(url));
-
-    request.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
-    if (!apiKey().isEmpty()) {
-        request.setRawHeader("Authorization", QString("Bearer %1").arg(apiKey()).toUtf8());
-    }
-
-    QNetworkReply *reply = manager.get(request);
-    QEventLoop loop;
-    QObject::connect(reply, &QNetworkReply::finished, &loop, &QEventLoop::quit);
-    loop.exec();
-
-    if (reply->error() == QNetworkReply::NoError) {
-        QByteArray responseData = reply->readAll();
-        QJsonDocument jsonResponse = QJsonDocument::fromJson(responseData);
-        QJsonObject jsonObject = jsonResponse.object();
-
-        if (jsonObject.contains("data") && jsonObject["object"].toString() == "list") {
-            QJsonArray modelArray = jsonObject["data"].toArray();
-            for (const QJsonValue &value : modelArray) {
-                QJsonObject modelObject = value.toObject();
-                if (modelObject.contains("id")) {
-                    QString modelId = modelObject["id"].toString();
-                    models.append(modelId);
-                }
-            }
-        }
-    } else {
-        LOG_MESSAGE(QString("Error fetching Mistral AI models: %1").arg(reply->errorString()));
-    }
-
-    reply->deleteLater();
-    return models;
-}
-
-QList<QString> MistralAIProvider::validateRequest(
-    const QJsonObject &request, LLMCore::TemplateType type)
-{
-    const auto fimReq = QJsonObject{
-        {"model", {}},
-        {"max_tokens", {}},
-        {"stream", {}},
-        {"temperature", {}},
-        {"prompt", {}},
-        {"suffix", {}}};
-
-    const auto templateReq = QJsonObject{
-        {"model", {}},
-        {"messages", QJsonArray{{QJsonObject{{"role", {}}, {"content", {}}}}}},
-        {"temperature", {}},
-        {"max_tokens", {}},
-        {"top_p", {}},
-        {"frequency_penalty", {}},
-        {"presence_penalty", {}},
-        {"stop", QJsonArray{}},
-        {"stream", {}}};
-
-    return LLMCore::ValidationUtils::validateRequestFields(
-        request, type == LLMCore::TemplateType::FIM ? fimReq : templateReq);
-}
-
-QString MistralAIProvider::apiKey() const
-{
-    return Settings::providerSettings().mistralAiApiKey();
-}
-
-void MistralAIProvider::prepareNetworkRequest(QNetworkRequest &networkRequest) const
-{
-    networkRequest.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
-
-    if (!apiKey().isEmpty()) {
-        networkRequest.setRawHeader("Authorization", QString("Bearer %1").arg(apiKey()).toUtf8());
-    }
-}
-
-LLMCore::ProviderID MistralAIProvider::providerID() const
-{
-    return LLMCore::ProviderID::MistralAI;
-}
-
-void MistralAIProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
-{
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
-
-    if (type == LLMCore::RequestType::Chat) {
-        auto &settings = Settings::chatAssistantSettings();
-
-        request["max_tokens"] = settings.maxTokens();
-        request["temperature"] = settings.temperature();
-
-        if (settings.useTopP())
-            request["top_p"] = settings.topP();
-
-        // request["random_seed"] = "";
-
-        if (settings.useFrequencyPenalty())
-            request["frequency_penalty"] = settings.frequencyPenalty();
-        if (settings.usePresencePenalty())
-            request["presence_penalty"] = settings.presencePenalty();
-
-    } else {
-        auto &settings = Settings::codeCompletionSettings();
-
-        request["max_tokens"] = settings.maxTokens();
-        request["temperature"] = settings.temperature();
-
-        if (settings.useTopP())
-            request["top_p"] = settings.topP();
-
-        // request["random_seed"] = "";
-    }
-}
-
-} // namespace QodeAssist::Providers

providers/MistralAIProvider.hpp

@@ -1,47 +0,0 @@
-/* 
- * Copyright (C) 2024 Petr Mironychev
- *
- * This file is part of QodeAssist.
- *
- * QodeAssist is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * QodeAssist is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#pragma once
-
-#include "llmcore/Provider.hpp"
-
-namespace QodeAssist::Providers {
-
-class MistralAIProvider : public LLMCore::Provider
-{
-public:
-    QString name() const override;
-    QString url() const override;
-    QString completionEndpoint() const override;
-    QString chatEndpoint() const override;
-    bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
-    bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
-    QList<QString> getInstalledModels(const QString &url) override;
-    QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
-    QString apiKey() const override;
-    void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
-};
-
-} // namespace QodeAssist::Providers

providers/OllamaProvider.cpp

@@ -33,6 +33,8 @@
 
 namespace QodeAssist::Providers {
 
+OllamaProvider::OllamaProvider() {}
+
 QString OllamaProvider::name() const
 {
     return "Ollama";
@@ -58,18 +60,8 @@ bool OllamaProvider::supportsModelListing() const
     return true;
 }
 
-void OllamaProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
+void OllamaProvider::prepareRequest(QJsonObject &request, LLMCore::RequestType type)
 {
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
-
     auto applySettings = [&request](const auto &settings) {
         QJsonObject options;
         options["num_predict"] = settings.maxTokens();
@@ -200,7 +192,7 @@ QList<QString> OllamaProvider::validateRequest(const QJsonObject &request, LLMCo
              {"presence_penalty", {}}}}};
 
     return LLMCore::ValidationUtils::validateRequestFields(
-        request, type == LLMCore::TemplateType::FIM ? fimReq : messageReq);
+        request, type == LLMCore::TemplateType::Fim ? fimReq : messageReq);
 }
 
 QString OllamaProvider::apiKey() const
@@ -211,11 +203,6 @@ QString OllamaProvider::apiKey() const
 void OllamaProvider::prepareNetworkRequest(QNetworkRequest &networkRequest) const
 {
     networkRequest.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
-}
-
-LLMCore::ProviderID OllamaProvider::providerID() const
-{
-    return LLMCore::ProviderID::Ollama;
-}
+};
 
 } // namespace QodeAssist::Providers

providers/OllamaProvider.hpp

@@ -26,22 +26,19 @@ namespace QodeAssist::Providers {
 class OllamaProvider : public LLMCore::Provider
 {
 public:
+    OllamaProvider();
+
     QString name() const override;
     QString url() const override;
     QString completionEndpoint() const override;
     QString chatEndpoint() const override;
     bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
+    void prepareRequest(QJsonObject &request, LLMCore::RequestType type) override;
     bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
     QList<QString> getInstalledModels(const QString &url) override;
     QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
     QString apiKey() const override;
     void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
 };
 
 } // namespace QodeAssist::Providers

providers/OpenAICompatProvider.cpp

@@ -34,6 +34,8 @@
 
 namespace QodeAssist::Providers {
 
+OpenAICompatProvider::OpenAICompatProvider() {}
+
 QString OpenAICompatProvider::name() const
 {
     return "OpenAI Compatible";
@@ -59,17 +61,20 @@ bool OpenAICompatProvider::supportsModelListing() const
     return false;
 }
 
-void OpenAICompatProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
+void OpenAICompatProvider::prepareRequest(QJsonObject &request, LLMCore::RequestType type)
 {
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
+    auto prepareMessages = [](QJsonObject &req) -> QJsonArray {
+        QJsonArray messages;
+        if (req.contains("system")) {
+            messages.append(
+                QJsonObject{{"role", "system"}, {"content", req.take("system").toString()}});
+        }
+        if (req.contains("prompt")) {
+            messages.append(
+                QJsonObject{{"role", "user"}, {"content", req.take("prompt").toString()}});
+        }
+        return messages;
+    };
 
     auto applyModelParams = [&request](const auto &settings) {
         request["max_tokens"] = settings.maxTokens();
@@ -85,6 +90,11 @@ void OpenAICompatProvider::prepareRequest(
             request["presence_penalty"] = settings.presencePenalty();
     };
 
+    QJsonArray messages = prepareMessages(request);
+    if (!messages.isEmpty()) {
+        request["messages"] = std::move(messages);
+    }
+
     if (type == LLMCore::RequestType::CodeCompletion) {
         applyModelParams(Settings::codeCompletionSettings());
     } else {
@@ -180,9 +190,4 @@ void OpenAICompatProvider::prepareNetworkRequest(QNetworkRequest &networkRequest
     }
 }
 
-LLMCore::ProviderID OpenAICompatProvider::providerID() const
-{
-    return LLMCore::ProviderID::OpenAICompatible;
-}
-
 } // namespace QodeAssist::Providers

providers/OpenAICompatProvider.hpp

@@ -26,22 +26,19 @@ namespace QodeAssist::Providers {
 class OpenAICompatProvider : public LLMCore::Provider
 {
 public:
+    OpenAICompatProvider();
+
     QString name() const override;
     QString url() const override;
     QString completionEndpoint() const override;
     QString chatEndpoint() const override;
     bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
+    void prepareRequest(QJsonObject &request, LLMCore::RequestType type) override;
     bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
     QList<QString> getInstalledModels(const QString &url) override;
     QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
     QString apiKey() const override;
     void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
 };
 
 } // namespace QodeAssist::Providers

providers/OpenAIProvider.cpp

@@ -35,6 +35,8 @@
 
 namespace QodeAssist::Providers {
 
+OpenAIProvider::OpenAIProvider() {}
+
 QString OpenAIProvider::name() const
 {
     return "OpenAI";
@@ -60,17 +62,20 @@ bool OpenAIProvider::supportsModelListing() const
     return true;
 }
 
-void OpenAIProvider::prepareRequest(
-    QJsonObject &request,
-    LLMCore::PromptTemplate *prompt,
-    LLMCore::ContextData context,
-    LLMCore::RequestType type)
+void OpenAIProvider::prepareRequest(QJsonObject &request, LLMCore::RequestType type)
 {
-    if (!prompt->isSupportProvider(providerID())) {
-        LOG_MESSAGE(QString("Template %1 doesn't support %2 provider").arg(name(), prompt->name()));
-    }
-
-    prompt->prepareRequest(request, context);
+    auto prepareMessages = [](QJsonObject &req) -> QJsonArray {
+        QJsonArray messages;
+        if (req.contains("system")) {
+            messages.append(
+                QJsonObject{{"role", "system"}, {"content", req.take("system").toString()}});
+        }
+        if (req.contains("prompt")) {
+            messages.append(
+                QJsonObject{{"role", "user"}, {"content", req.take("prompt").toString()}});
+        }
+        return messages;
+    };
 
     auto applyModelParams = [&request](const auto &settings) {
         request["max_tokens"] = settings.maxTokens();
@@ -86,6 +91,11 @@ void OpenAIProvider::prepareRequest(
             request["presence_penalty"] = settings.presencePenalty();
     };
 
+    QJsonArray messages = prepareMessages(request);
+    if (!messages.isEmpty()) {
+        request["messages"] = std::move(messages);
+    }
+
     if (type == LLMCore::RequestType::CodeCompletion) {
         applyModelParams(Settings::codeCompletionSettings());
     } else {
@@ -216,9 +226,4 @@ void OpenAIProvider::prepareNetworkRequest(QNetworkRequest &networkRequest) cons
     }
 }
 
-LLMCore::ProviderID OpenAIProvider::providerID() const
-{
-    return LLMCore::ProviderID::OpenAI;
-}
-
 } // namespace QodeAssist::Providers

providers/OpenAIProvider.hpp

@@ -26,22 +26,19 @@ namespace QodeAssist::Providers {
 class OpenAIProvider : public LLMCore::Provider
 {
 public:
+    OpenAIProvider();
+
     QString name() const override;
     QString url() const override;
     QString completionEndpoint() const override;
     QString chatEndpoint() const override;
     bool supportsModelListing() const override;
-    void prepareRequest(
-        QJsonObject &request,
-        LLMCore::PromptTemplate *prompt,
-        LLMCore::ContextData context,
-        LLMCore::RequestType type) override;
+    void prepareRequest(QJsonObject &request, LLMCore::RequestType type) override;
     bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
     QList<QString> getInstalledModels(const QString &url) override;
     QList<QString> validateRequest(const QJsonObject &request, LLMCore::TemplateType type) override;
     QString apiKey() const override;
     void prepareNetworkRequest(QNetworkRequest &networkRequest) const override;
-    LLMCore::ProviderID providerID() const override;
 };
 
 } // namespace QodeAssist::Providers

providers/OpenRouterAIProvider.cpp

@@ -19,6 +19,8 @@
 
 #include "OpenRouterAIProvider.hpp"
 
+#include "settings/ChatAssistantSettings.hpp"
+#include "settings/CodeCompletionSettings.hpp"
 #include "settings/ProviderSettings.hpp"
 
 #include <QJsonArray>
@@ -31,6 +33,8 @@
 
 namespace QodeAssist::Providers {
 
+OpenRouterProvider::OpenRouterProvider() {}
+
 QString OpenRouterProvider::name() const
 {
     return "OpenRouter";
@@ -41,6 +45,47 @@ QString OpenRouterProvider::url() const
     return "https://openrouter.ai/api";
 }
 
+void OpenRouterProvider::prepareRequest(QJsonObject &request, LLMCore::RequestType type)
+{
+    auto prepareMessages = [](QJsonObject &req) -> QJsonArray {
+        QJsonArray messages;
+        if (req.contains("system")) {
+            messages.append(
+                QJsonObject{{"role", "system"}, {"content", req.take("system").toString()}});
+        }
+        if (req.contains("prompt")) {
+            messages.append(
+                QJsonObject{{"role", "user"}, {"content", req.take("prompt").toString()}});
+        }
+        return messages;
+    };
+
+    auto applyModelParams = [&request](const auto &settings) {
+        request["max_tokens"] = settings.maxTokens();
+        request["temperature"] = settings.temperature();
+
+        if (settings.useTopP())
+            request["top_p"] = settings.topP();
+        if (settings.useTopK())
+            request["top_k"] = settings.topK();
+        if (settings.useFrequencyPenalty())
+            request["frequency_penalty"] = settings.frequencyPenalty();
+        if (settings.usePresencePenalty())
+            request["presence_penalty"] = settings.presencePenalty();
+    };
+
+    QJsonArray messages = prepareMessages(request);
+    if (!messages.isEmpty()) {
+        request["messages"] = std::move(messages);
+    }
+
+    if (type == LLMCore::RequestType::CodeCompletion) {
+        applyModelParams(Settings::codeCompletionSettings());
+    } else {
+        applyModelParams(Settings::chatAssistantSettings());
+    }
+}
+
 bool OpenRouterProvider::handleResponse(QNetworkReply *reply, QString &accumulatedResponse)
 {
     QByteArray data = reply->readAll();
@@ -97,9 +142,4 @@ QString OpenRouterProvider::apiKey() const
     return Settings::providerSettings().openRouterApiKey();
 }
 
-LLMCore::ProviderID OpenRouterProvider::providerID() const
-{
-    return LLMCore::ProviderID::OpenRouter;
-}
-
 } // namespace QodeAssist::Providers

providers/OpenRouterAIProvider.hpp

@@ -27,11 +27,13 @@ namespace QodeAssist::Providers {
 class OpenRouterProvider : public OpenAICompatProvider
 {
 public:
+    OpenRouterProvider();
+
     QString name() const override;
     QString url() const override;
+    void prepareRequest(QJsonObject &request, LLMCore::RequestType type) override;
     bool handleResponse(QNetworkReply *reply, QString &accumulatedResponse) override;
     QString apiKey() const override;
-    LLMCore::ProviderID providerID() const override;
 };
 
 } // namespace QodeAssist::Providers

providers/Providers.hpp

@@ -21,9 +21,7 @@
 
 #include "llmcore/ProvidersManager.hpp"
 #include "providers/ClaudeProvider.hpp"
-#include "providers/GoogleAIProvider.hpp"
 #include "providers/LMStudioProvider.hpp"
-#include "providers/MistralAIProvider.hpp"
 #include "providers/OllamaProvider.hpp"
 #include "providers/OpenAICompatProvider.hpp"
 #include "providers/OpenAIProvider.hpp"
@@ -35,13 +33,11 @@ inline void registerProviders()
 {
     auto &providerManager = LLMCore::ProvidersManager::instance();
     providerManager.registerProvider<OllamaProvider>();
+    providerManager.registerProvider<LMStudioProvider>();
+    providerManager.registerProvider<OpenAICompatProvider>();
+    providerManager.registerProvider<OpenRouterProvider>();
     providerManager.registerProvider<ClaudeProvider>();
     providerManager.registerProvider<OpenAIProvider>();
-    providerManager.registerProvider<OpenAICompatProvider>();
-    providerManager.registerProvider<LMStudioProvider>();
-    providerManager.registerProvider<OpenRouterProvider>();
-    providerManager.registerProvider<MistralAIProvider>();
-    providerManager.registerProvider<GoogleAIProvider>();
 }
 
 } // namespace QodeAssist::Providers

qodeassist.cpp

@@ -108,10 +108,8 @@ public:
             UpdateDialog::checkForUpdatesAndShow(Core::ICore::mainWindow());
         });
 
-        if (Settings::generalSettings().enableChat()) {
-            m_chatOutputPane = new Chat::ChatOutputPane(this);
-            m_navigationPanel = new Chat::NavigationPanel();
-        }
+        m_chatOutputPane = new Chat::ChatOutputPane(this);
+        m_navigationPanel = new Chat::NavigationPanel();
 
         Settings::setupProjectPanel();
         ConfigurationManager::instance().init();

settings/CodeCompletionSettings.cpp

@@ -90,7 +90,7 @@ CodeCompletionSettings::CodeCompletionSettings()
     maxTokens.setSettingsKey(Constants::CC_MAX_TOKENS);
     maxTokens.setLabelText(Tr::tr("Max Tokens:"));
     maxTokens.setRange(-1, 900000);
-    maxTokens.setDefaultValue(100);
+    maxTokens.setDefaultValue(50);
 
     // Advanced Parameters
     useTopP.setSettingsKey(Constants::CC_USE_TOP_P);
@@ -151,47 +151,31 @@ CodeCompletionSettings::CodeCompletionSettings()
 
     systemPrompt.setSettingsKey(Constants::CC_SYSTEM_PROMPT);
     systemPrompt.setDisplayStyle(Utils::StringAspect::TextEditDisplay);
-    systemPromptForNonFimModels.setLabelText(Tr::tr("System prompt for FIM models:"));
     systemPrompt.setDefaultValue(
-        "You are an expert C++, Qt, and QML code completion assistant. Your task is to provide "
-        "precise and contextually appropriate code completions.\n\n");
+        "You are an expert in C++, Qt, and QML programming. Your task is to provide code "
+        "suggestions that seamlessly integrate with existing code. Do not repeat code from position "
+        "before or after <cursor>. You will receive a code context with specified insertion points. "
+        "Your goal is to complete only one code block."
+        "Here is the code context with insertion points:<code_context>Before: {{variable}}After: "
+        "{{variable}}</code_context> Instructions: 1. Carefully analyze the provided code context. "
+        "2. Consider the existing code and the specified insertion points.3. Generate a code "
+        "suggestion that completes one logic expression between the 'Before' and 'After' points. "
+        "4. Ensure your suggestion does not repeat any existing code. 5. Format your suggestion as "
+        "a code block using triple backticks. 6. Do not include any comments or descriptions with "
+        "your code suggestion.");
 
     useUserMessageTemplateForCC.setSettingsKey(Constants::CC_USE_USER_TEMPLATE);
     useUserMessageTemplateForCC.setDefaultValue(true);
-    useUserMessageTemplateForCC.setLabelText(
-        Tr::tr("Use special system prompt and user message for non FIM models"));
-
-    systemPromptForNonFimModels.setSettingsKey(Constants::CC_SYSTEM_PROMPT_FOR_NON_FIM);
-    systemPromptForNonFimModels.setDisplayStyle(Utils::StringAspect::TextEditDisplay);
-    systemPromptForNonFimModels.setLabelText(Tr::tr("System prompt for non FIM models:"));
-    systemPromptForNonFimModels.setDefaultValue(
-        "You are an expert C++, Qt, and QML code completion assistant. Your task is to provide "
-        "precise and contextually appropriate code completions.\n\n"
-        "Core Requirements:\n"
-        "1. Continue code exactly from the cursor position, ensuring it properly connects with any "
-        "existing code after the cursor\n"
-        "2. Never repeat existing code before or after the cursor\n"
-        "Specific Guidelines:\n"
-        "- For function calls: Complete parameters with appropriate types and names\n"
-        "- For class members: Respect access modifiers and class conventions\n"
-        "- Respect existing indentation and formatting\n"
-        "- Consider scope and visibility of referenced symbols\n"
-        "- Ensure seamless integration with code both before and after the cursor\n\n"
-        "Context Format:\n"
-        "<code_context>\n"
-        "{{code before cursor}}<cursor>{{code after cursor}}\n"
-        "</code_context>\n\n"
-        "Response Format:\n"
-        "- No explanations or comments\n"
-        "- Only include new characters needed to create valid code\n"
-        "- Should be codeblock with language\n");
+    useUserMessageTemplateForCC.setLabelText(Tr::tr("Use User Template for code completion message for non-FIM models"));
 
     userMessageTemplateForCC.setSettingsKey(Constants::CC_USER_TEMPLATE);
     userMessageTemplateForCC.setDisplayStyle(Utils::StringAspect::TextEditDisplay);
-    userMessageTemplateForCC.setLabelText(Tr::tr("User message for non FIM models:"));
-    userMessageTemplateForCC.setDefaultValue(
-        "Here is the code context with insertion points:\n"
-        "<code_context>\n${prefix}<cursor>${suffix}\n</code_context>\n\n");
+    userMessageTemplateForCC.setDefaultValue("Here is the code context with insertion points: <code_context>"
+                                             "\nBefore: %1After: %2\n </code_context>");
+
+    useFilePathInContext.setSettingsKey(Constants::CC_USE_FILE_PATH_IN_CONTEXT);
+    useFilePathInContext.setDefaultValue(true);
+    useFilePathInContext.setLabelText(Tr::tr("Use File Path in Context"));
 
     useProjectChangesCache.setSettingsKey(Constants::CC_USE_PROJECT_CHANGES_CACHE);
     useProjectChangesCache.setDefaultValue(true);
@@ -252,14 +236,10 @@ CodeCompletionSettings::CodeCompletionSettings()
 
         auto contextItem = Column{Row{contextGrid, Stretch{1}},
                                   Row{useSystemPrompt, Stretch{1}},
-                                  Group{title(Tr::tr("Prompts for FIM models")),
-                                        Column{systemPrompt}},
-                                  Group{title(Tr::tr("Prompts for Non FIM models")),
-                                        Column{
-                                            Row{useUserMessageTemplateForCC, Stretch{1}},
-                                            systemPromptForNonFimModels,
-                                            userMessageTemplateForCC,
-                                        }},
+                                  systemPrompt,
+                                  Row{useUserMessageTemplateForCC, Stretch{1}},
+                                  userMessageTemplateForCC,
+                                  Row{useFilePathInContext, Stretch{1}},
                                   Row{useProjectChangesCache, maxChangesCacheSize, Stretch{1}}};
 
         return Column{
@@ -347,24 +327,14 @@ void CodeCompletionSettings::resetSettingsToDefaults()
         resetAspect(readStringsAfterCursor);
         resetAspect(useSystemPrompt);
         resetAspect(systemPrompt);
+        resetAspect(useFilePathInContext);
         resetAspect(useProjectChangesCache);
         resetAspect(maxChangesCacheSize);
         resetAspect(ollamaLivetime);
         resetAspect(contextWindow);
-        resetAspect(useUserMessageTemplateForCC);
-        resetAspect(userMessageTemplateForCC);
-        resetAspect(systemPromptForNonFimModels);
     }
 }
 
-QString CodeCompletionSettings::processMessageToFIM(const QString &prefix, const QString &suffix)
-{
-    QString result = userMessageTemplateForCC();
-    result.replace("${prefix}", prefix);
-    result.replace("${suffix}", suffix);
-    return result;
-}
-
 class CodeCompletionSettingsPage : public Core::IOptionsPage
 {
 public:

settings/CodeCompletionSettings.hpp

@@ -67,8 +67,8 @@ public:
     Utils::BoolAspect useSystemPrompt{this};
     Utils::StringAspect systemPrompt{this};
     Utils::BoolAspect useUserMessageTemplateForCC{this};
-    Utils::StringAspect systemPromptForNonFimModels{this};
     Utils::StringAspect userMessageTemplateForCC{this};
+    Utils::BoolAspect useFilePathInContext{this};
     Utils::BoolAspect useProjectChangesCache{this};
     Utils::IntegerAspect maxChangesCacheSize{this};
 
@@ -79,8 +79,6 @@ public:
     // API Configuration Settings
     Utils::StringAspect apiKey{this};
 
-    QString processMessageToFIM(const QString &prefix, const QString &suffix);
-
 private:
     void setupConnections();
     void resetSettingsToDefaults();

settings/GeneralSettings.cpp

@@ -21,7 +21,6 @@
 
 #include <coreplugin/dialogs/ioptionspage.h>
 #include <coreplugin/icore.h>
-#include <utils/detailswidget.h>
 #include <utils/layoutbuilder.h>
 #include <utils/utilsicons.h>
 #include <QInputDialog>
@@ -66,10 +65,6 @@ GeneralSettings::GeneralSettings()
     enableCheckUpdate.setLabelText(TrConstants::ENABLE_CHECK_UPDATE_ON_START);
     enableCheckUpdate.setDefaultValue(true);
 
-    enableChat.setSettingsKey(Constants::ENABLE_CHAT);
-    enableChat.setLabelText(TrConstants::ENABLE_CHAT);
-    enableChat.setDefaultValue(true);
-
     resetToDefaults.m_buttonText = TrConstants::RESET_TO_DEFAULTS;
     checkUpdate.m_buttonText = TrConstants::CHECK_UPDATE;
 
@@ -94,11 +89,6 @@ GeneralSettings::GeneralSettings()
     ccStatus.setDefaultValue("");
     ccTest.m_buttonText = TrConstants::TEST;
 
-    ccTemplateDescription.setDisplayStyle(Utils::StringAspect::TextEditDisplay);
-    ccTemplateDescription.setReadOnly(true);
-    ccTemplateDescription.setDefaultValue("");
-    ccTemplateDescription.setLabelText(TrConstants::CURRENT_TEMPLATE_DESCRIPTION);
-
     // preset1
     specifyPreset1.setSettingsKey(Constants::CC_SPECIFY_PRESET1);
     specifyPreset1.setLabelText(TrConstants::ADD_NEW_PRESET);
@@ -154,11 +144,6 @@ GeneralSettings::GeneralSettings()
     caStatus.setDefaultValue("");
     caTest.m_buttonText = TrConstants::TEST;
 
-    caTemplateDescription.setDisplayStyle(Utils::StringAspect::TextEditDisplay);
-    caTemplateDescription.setReadOnly(true);
-    caTemplateDescription.setDefaultValue("");
-    caTemplateDescription.setLabelText(TrConstants::CURRENT_TEMPLATE_DESCRIPTION);
-
     readSettings();
 
     Logger::instance().setLoggingEnabled(enableLogging());
@@ -190,19 +175,13 @@ GeneralSettings::GeneralSettings()
 
         auto ccGroup = Group{
             title(TrConstants::CODE_COMPLETION),
-            Column{
-                ccGrid,
-                ccTemplateDescription,
-                Row{specifyPreset1, preset1Language, Stretch{1}},
-                ccPreset1Grid}};
-        auto caGroup
-            = Group{title(TrConstants::CHAT_ASSISTANT), Column{caGrid, caTemplateDescription}};
+            Column{ccGrid, Row{specifyPreset1, preset1Language, Stretch{1}}, ccPreset1Grid}};
+        auto caGroup = Group{title(TrConstants::CHAT_ASSISTANT), caGrid};
 
         auto rootLayout = Column{
             Row{enableQodeAssist, Stretch{1}, Row{checkUpdate, resetToDefaults}},
             Row{enableLogging, Stretch{1}},
             Row{enableCheckUpdate, Stretch{1}},
-            Row{enableChat, Stretch{1}},
             Space{8},
             ccGroup,
             Space{8},
@@ -213,8 +192,10 @@ GeneralSettings::GeneralSettings()
     });
 }
 
-void GeneralSettings::showSelectionDialog(
-    const QStringList &data, Utils::StringAspect &aspect, const QString &title, const QString &text)
+void GeneralSettings::showSelectionDialog(const QStringList &data,
+                                          Utils::StringAspect &aspect,
+                                          const QString &title,
+                                          const QString &text)
 {
     if (data.isEmpty())
         return;
@@ -258,7 +239,6 @@ void GeneralSettings::showModelsNotFoundDialog(Utils::StringAspect &aspect)
         auto selectProviderBtn = new QPushButton(TrConstants::SELECT_PROVIDER);
         auto selectUrlBtn = new QPushButton(TrConstants::SELECT_URL);
         auto enterManuallyBtn = new QPushButton(TrConstants::ENTER_MODEL_MANUALLY);
-        auto configureApiKeyBtn = new QPushButton(TrConstants::CONFIGURE_API_KEY);
 
         connect(selectProviderBtn, &QPushButton::clicked, &dialog, [this, providerButton, &dialog]() {
             dialog.close();
@@ -275,15 +255,9 @@ void GeneralSettings::showModelsNotFoundDialog(Utils::StringAspect &aspect)
             showModelsNotSupportedDialog(aspect);
         });
 
-        connect(configureApiKeyBtn, &QPushButton::clicked, &dialog, [&dialog]() {
-            dialog.close();
-            Core::ICore::showOptionsDialog(Constants::QODE_ASSIST_PROVIDER_SETTINGS_PAGE_ID);
-        });
-
         dialog.buttonLayout()->addWidget(selectProviderBtn);
         dialog.buttonLayout()->addWidget(selectUrlBtn);
         dialog.buttonLayout()->addWidget(enterManuallyBtn);
-        dialog.buttonLayout()->addWidget(configureApiKeyBtn);
     }
 
     auto closeBtn = new QPushButton(TrConstants::CLOSE);
@@ -392,7 +366,7 @@ void GeneralSettings::setupConnections()
 
     connect(&specifyPreset1, &Utils::BoolAspect::volatileValueChanged, this, [this]() {
         updatePreset1Visiblity(specifyPreset1.volatileValue());
-    });    
+    });
 }
 
 void GeneralSettings::resetPageToDefaults()

settings/GeneralSettings.hpp

@@ -36,8 +36,6 @@ public:
     Utils::BoolAspect enableQodeAssist{this};
     Utils::BoolAspect enableLogging{this};
     Utils::BoolAspect enableCheckUpdate{this};
-    Utils::BoolAspect enableChat{this};
-
     ButtonAspect checkUpdate{this};
     ButtonAspect resetToDefaults{this};
 
@@ -57,8 +55,6 @@ public:
     Utils::StringAspect ccStatus{this};
     ButtonAspect ccTest{this};
 
-    Utils::StringAspect ccTemplateDescription{this};
-
     // TODO create dynamic presets system
     // preset1 for code completion settings
     Utils::BoolAspect specifyPreset1{this};
@@ -92,8 +88,6 @@ public:
     Utils::StringAspect caStatus{this};
     ButtonAspect caTest{this};
 
-    Utils::StringAspect caTemplateDescription{this};
-
     void showSelectionDialog(const QStringList &data,
                              Utils::StringAspect &aspect,
                              const QString &title = {},

settings/ProviderSettings.cpp

@@ -69,7 +69,6 @@ ProviderSettings::ProviderSettings()
     claudeApiKey.setDefaultValue("");
     claudeApiKey.setAutoApply(true);
 
-    // OpenAI Settings
     openAiApiKey.setSettingsKey(Constants::OPEN_AI_API_KEY);
     openAiApiKey.setLabelText(Tr::tr("OpenAI API Key:"));
     openAiApiKey.setDisplayStyle(Utils::StringAspect::LineEditDisplay);
@@ -78,24 +77,6 @@ ProviderSettings::ProviderSettings()
     openAiApiKey.setDefaultValue("");
     openAiApiKey.setAutoApply(true);
 
-    // MistralAI Settings
-    mistralAiApiKey.setSettingsKey(Constants::MISTRAL_AI_API_KEY);
-    mistralAiApiKey.setLabelText(Tr::tr("Mistral AI API Key:"));
-    mistralAiApiKey.setDisplayStyle(Utils::StringAspect::LineEditDisplay);
-    mistralAiApiKey.setPlaceHolderText(Tr::tr("Enter your API key here"));
-    mistralAiApiKey.setHistoryCompleter(Constants::MISTRAL_AI_API_KEY_HISTORY);
-    mistralAiApiKey.setDefaultValue("");
-    mistralAiApiKey.setAutoApply(true);
-
-    // GoogleAI Settings
-    googleAiApiKey.setSettingsKey(Constants::GOOGLE_AI_API_KEY);
-    googleAiApiKey.setLabelText(Tr::tr("Google AI API Key:"));
-    googleAiApiKey.setDisplayStyle(Utils::StringAspect::LineEditDisplay);
-    googleAiApiKey.setPlaceHolderText(Tr::tr("Enter your API key here"));
-    googleAiApiKey.setHistoryCompleter(Constants::GOOGLE_AI_API_KEY_HISTORY);
-    googleAiApiKey.setDefaultValue("");
-    googleAiApiKey.setAutoApply(true);
-
     resetToDefaults.m_buttonText = Tr::tr("Reset Page to Defaults");
 
     readSettings();
@@ -115,10 +96,6 @@ ProviderSettings::ProviderSettings()
             Group{title(Tr::tr("OpenAI Compatible Settings")), Column{openAiCompatApiKey}},
             Space{8},
             Group{title(Tr::tr("Claude Settings")), Column{claudeApiKey}},
-            Space{8},
-            Group{title(Tr::tr("Mistral AI Settings")), Column{mistralAiApiKey}},
-            Space{8},
-            Group{title(Tr::tr("Google AI Settings")), Column{googleAiApiKey}},
             Stretch{1}};
     });
 }
@@ -135,12 +112,6 @@ void ProviderSettings::setupConnections()
     });
     connect(&claudeApiKey, &ButtonAspect::changed, this, [this]() { claudeApiKey.writeSettings(); });
     connect(&openAiApiKey, &ButtonAspect::changed, this, [this]() { openAiApiKey.writeSettings(); });
-    connect(&mistralAiApiKey, &ButtonAspect::changed, this, [this]() {
-        mistralAiApiKey.writeSettings();
-    });
-    connect(&googleAiApiKey, &ButtonAspect::changed, this, [this]() {
-        googleAiApiKey.writeSettings();
-    });
 }
 
 void ProviderSettings::resetSettingsToDefaults()
@@ -157,8 +128,6 @@ void ProviderSettings::resetSettingsToDefaults()
         resetAspect(openAiCompatApiKey);
         resetAspect(claudeApiKey);
         resetAspect(openAiApiKey);
-        resetAspect(mistralAiApiKey);
-        resetAspect(googleAiApiKey);
     }
 }
 

settings/ProviderSettings.hpp

@@ -37,8 +37,6 @@ public:
     Utils::StringAspect openAiCompatApiKey{this};
     Utils::StringAspect claudeApiKey{this};
     Utils::StringAspect openAiApiKey{this};
-    Utils::StringAspect mistralAiApiKey{this};
-    Utils::StringAspect googleAiApiKey{this};
 
 private:
     void setupConnections();

settings/SettingsConstants.hpp

@@ -60,7 +60,6 @@ const char ENABLE_QODE_ASSIST[] = "QodeAssist.enableQodeAssist";
 const char CC_AUTO_COMPLETION[] = "QodeAssist.ccAutoCompletion";
 const char ENABLE_LOGGING[] = "QodeAssist.enableLogging";
 const char ENABLE_CHECK_UPDATE[] = "QodeAssist.enableCheckUpdate";
-const char ENABLE_CHAT[] = "QodeAssist.enableChat";
 
 const char PROVIDER_PATHS[] = "QodeAssist.providerPaths";
 const char СС_START_SUGGESTION_TIMER[] = "QodeAssist.startSuggestionTimer";
@@ -101,18 +100,14 @@ const char CLAUDE_API_KEY[] = "QodeAssist.claudeApiKey";
 const char CLAUDE_API_KEY_HISTORY[] = "QodeAssist.claudeApiKeyHistory";
 const char OPEN_AI_API_KEY[] = "QodeAssist.openAiApiKey";
 const char OPEN_AI_API_KEY_HISTORY[] = "QodeAssist.openAiApiKeyHistory";
-const char MISTRAL_AI_API_KEY[] = "QodeAssist.mistralAiApiKey";
-const char MISTRAL_AI_API_KEY_HISTORY[] = "QodeAssist.mistralAiApiKeyHistory";
-const char GOOGLE_AI_API_KEY[] = "QodeAssist.googleAiApiKey";
-const char GOOGLE_AI_API_KEY_HISTORY[] = "QodeAssist.googleAiApiKeyHistory";
 
 // context settings
 const char CC_READ_FULL_FILE[] = "QodeAssist.ccReadFullFile";
 const char CC_READ_STRINGS_BEFORE_CURSOR[] = "QodeAssist.ccReadStringsBeforeCursor";
 const char CC_READ_STRINGS_AFTER_CURSOR[] = "QodeAssist.ccReadStringsAfterCursor";
 const char CC_USE_SYSTEM_PROMPT[] = "QodeAssist.ccUseSystemPrompt";
+const char CC_USE_FILE_PATH_IN_CONTEXT[] = "QodeAssist.ccUseFilePathInContext";
 const char CC_SYSTEM_PROMPT[] = "QodeAssist.ccSystemPrompt";
-const char CC_SYSTEM_PROMPT_FOR_NON_FIM[] = "QodeAssist.ccSystemPromptForNonFim";
 const char CC_USE_USER_TEMPLATE[] = "QodeAssist.ccUseUserTemplate";
 const char CC_USER_TEMPLATE[] = "QodeAssist.ccUserTemplate";
 const char CC_USE_PROJECT_CHANGES_CACHE[] = "QodeAssist.ccUseProjectChangesCache";

settings/SettingsTr.hpp

@@ -39,9 +39,6 @@ inline const char *TEST = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Test");
 inline const char *ENABLE_LOG = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Enable Logging");
 inline const char *ENABLE_CHECK_UPDATE_ON_START
     = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Check for updates when Qt Creator starts");
-inline const char *ENABLE_CHAT = QT_TRANSLATE_NOOP(
-    "QtC::QodeAssist",
-    "Enable Chat(If you have performance issues try disabling this, need restart QtC)");
 
 inline const char *CODE_COMPLETION = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Code Completion");
 inline const char *CHAT_ASSISTANT = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Chat Assistant");
@@ -49,8 +46,6 @@ inline const char *RESET_SETTINGS = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Reset
 inline const char *CONFIRMATION
     = QT_TRANSLATE_NOOP("QtC::QodeAssist",
                         "Are you sure you want to reset all settings to default values?");
-inline const char *CURRENT_TEMPLATE_DESCRIPTION
-    = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Current template description:");
 
 inline const char CONNECTION_ERROR[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Connection Error");
 inline const char NO_MODELS_FOUND[]
@@ -60,8 +55,7 @@ inline const char CHECK_CONNECTION[] = QT_TRANSLATE_NOOP(
     "Please verify the following:\n"
     "- Server is running and accessible\n"
     "- URL is correct\n"
-    "- Provider is properly configured\n"
-    "- API key is correctly set (if required)\n\n"
+    "- Provider is properly configured\n\n"
     "You can try selecting a different provider or changing the URL:");
 inline const char SELECT_PROVIDER[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Select Provider");
 inline const char SELECT_URL[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Select URL");
@@ -79,7 +73,6 @@ inline const char OK[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "OK");
 inline const char CANCEL[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Cancel");
 inline const char ENTER_MODEL_MANUALLY[]
     = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Enter Model Manually");
-inline const char CONFIGURE_API_KEY[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "Configure API Key");
 inline const char URL_SELECTION[] = QT_TRANSLATE_NOOP("QtC::QodeAssist", "URL Selection");
 inline const char URL_SELECTION_INFO[] = QT_TRANSLATE_NOOP(
     "QtC::QodeAssist",

templates/Alpaca.hpp

@@ -29,60 +29,38 @@ class Alpaca : public LLMCore::PromptTemplate
 public:
     QString name() const override { return "Alpaca"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override
     {
         return QStringList() << "### Instruction:" << "### Response:";
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QJsonArray messages;
+        QJsonArray messages = request["messages"].toArray();
 
-        QString fullContent;
+        for (int i = 0; i < messages.size(); ++i) {
+            QJsonObject message = messages[i].toObject();
+            QString role = message["role"].toString();
+            QString content = message["content"].toString();
 
-        if (context.systemPrompt) {
-            fullContent += context.systemPrompt.value() + "\n\n";
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                if (msg.role == "user") {
-                    fullContent += QString("### Instruction:\n%1\n\n").arg(msg.content);
-                } else if (msg.role == "assistant") {
-                    fullContent += QString("### Response:\n%1\n\n").arg(msg.content);
-                }
+            QString formattedContent;
+            if (role == "system") {
+                formattedContent = content + "\n\n";
+            } else if (role == "user") {
+                formattedContent = "### Instruction:\n" + content + "\n\n";
+            } else if (role == "assistant") {
+                formattedContent = "### Response:\n" + content + "\n\n";
             }
-        }
 
-        messages.append(QJsonObject{{"role", "user"}, {"content", fullContent}});
+            message["content"] = formattedContent;
+            messages[i] = message;
+        }
 
         request["messages"] = messages;
     }
     QString description() const override
     {
-        return "Template for models using Alpaca instruction format:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\n"
-               "      \"role\": \"user\",\n"
-               "      \"content\": \"<system prompt>\\n\\n"
-               "### Instruction:\\n<user message>\\n\\n"
-               "### Response:\\n<assistant response>\\n\\n\"\n"
-               "    }\n"
-               "  ]\n"
-               "}\n\n"
-               "Combines all messages into a single formatted prompt.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-        case QodeAssist::LLMCore::ProviderID::LMStudio:
-        case QodeAssist::LLMCore::ProviderID::OpenRouter:
-        case QodeAssist::LLMCore::ProviderID::OpenAICompatible:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: ### Instruction:\n### Response:\n";
     }
 };
 

templates/BasicChat.hpp

@@ -0,0 +1,40 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <QJsonArray>
+
+#include "llmcore/PromptTemplate.hpp"
+
+namespace QodeAssist::Templates {
+
+class BasicChat : public LLMCore::PromptTemplate
+{
+public:
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
+    QString name() const override { return "Basic Chat"; }
+    QString promptTemplate() const override { return {}; }
+    QStringList stopWords() const override { return QStringList(); }
+    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
+    {}
+    QString description() const override { return "chat without tokens"; }
+};
+
+} // namespace QodeAssist::Templates

templates/ChatML.hpp

@@ -30,60 +30,30 @@ class ChatML : public LLMCore::PromptTemplate
 public:
     QString name() const override { return "ChatML"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override
     {
         return QStringList() << "<|im_start|>" << "<|im_end|>";
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QJsonArray messages;
+        QJsonArray messages = request["messages"].toArray();
 
-        if (context.systemPrompt) {
-            messages.append(QJsonObject{
-                {"role", "system"},
-                {"content",
-                 QString("<|im_start|>system\n%2\n<|im_end|>").arg(context.systemPrompt.value())}});
-        }
+        for (int i = 0; i < messages.size(); ++i) {
+            QJsonObject message = messages[i].toObject();
+            QString role = message["role"].toString();
+            QString content = message["content"].toString();
 
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                messages.append(QJsonObject{
-                    {"role", msg.role},
-                    {"content",
-                     QString("<|im_start|>%1\n%2\n<|im_end|>").arg(msg.role, msg.content)}});
-            }
+            message["content"] = QString("<|im_start|>%1\n%2\n<|im_end|>").arg(role, content);
+
+            messages[i] = message;
         }
 
         request["messages"] = messages;
     }
     QString description() const override
     {
-        return "Template for models supporting ChatML format:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\n"
-               "      \"role\": \"system\",\n"
-               "      \"content\": \"<|im_start|>system\\n<system prompt>\\n<|im_end|>\"\n"
-               "    },\n"
-               "    {\n"
-               "      \"role\": \"user\",\n"
-               "      \"content\": \"<|im_start|>user\\n<user message>\\n<|im_end|>\"\n"
-               "    }\n"
-               "  ]\n"
-               "}\n\n"
-               "Compatible with multiple providers supporting the ChatML token format.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-        case QodeAssist::LLMCore::ProviderID::LMStudio:
-        case QodeAssist::LLMCore::ProviderID::OpenRouter:
-        case QodeAssist::LLMCore::ProviderID::OpenAICompatible:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: <|im_start|>%1\n%2\n<|im_end|>";
     }
 };
 

templates/Claude.hpp

@@ -30,46 +30,10 @@ class Claude : public LLMCore::PromptTemplate
 public:
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
     QString name() const override { return "Claude"; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList(); }
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        QJsonArray messages;
-
-        if (context.systemPrompt) {
-            request["system"] = context.systemPrompt.value();
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                if (msg.role != "system") {
-                    messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
-                }
-            }
-        }
-
-        request["messages"] = messages;
-    }
-    QString description() const override
-    {
-        return "Template for Anthropic's Claude models:\n\n"
-               "{\n"
-               "  \"system\": \"<system prompt>\",\n"
-               "  \"messages\": [\n"
-               "    {\"role\": \"user\", \"content\": \"<user message>\"},\n"
-               "    {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
-               "  ]\n"
-               "}\n\n"
-               "Formats content according to Claude API specifications.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Claude:
-            return true;
-        default:
-            return false;
-        }
-    }
+    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override {}
+    QString description() const override { return "Claude"; }
 };
 
 } // namespace QodeAssist::Templates

templates/CodeLlamaFim.hpp

@@ -26,35 +26,21 @@ namespace QodeAssist::Templates {
 class CodeLlamaFim : public LLMCore::PromptTemplate
 {
 public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::FIM; }
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Fim; }
     QString name() const override { return "CodeLlama FIM"; }
+    QString promptTemplate() const override { return "<PRE> %1 <SUF>%2 <MID>"; }
     QStringList stopWords() const override
     {
         return QStringList() << "<EOT>" << "<PRE>" << "<SUF" << "<MID>";
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        request["prompt"] = QString("<PRE> %1 <SUF>%2 <MID>")
-                                .arg(context.prefix.value_or(""), context.suffix.value_or(""));
-        request["system"] = context.systemPrompt.value_or("");
+        QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
+        request["prompt"] = formattedPrompt;
     }
     QString description() const override
     {
-        return "Specialized template for CodeLlama FIM:\n\n"
-               "{\n"
-               "  \"prompt\": \"<PRE> <code prefix> <SUF><code suffix> <MID>\",\n"
-               "  \"system\": \"<system prompt>\"\n"
-               "}\n\n"
-               "Optimized for code completion with CodeLlama models.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: <PRE> %1 <SUF>%2 <MID>";
     }
 };
 

templates/CodeLlamaQMLFim.hpp

@@ -26,8 +26,9 @@ namespace QodeAssist::Templates {
 class CodeLlamaQMLFim : public LLMCore::PromptTemplate
 {
 public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::FIM; }
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Fim; }
     QString name() const override { return "CodeLlama QML FIM"; }
+    QString promptTemplate() const override { return "<SUF>%1<PRE>%2<MID>"; }
     QStringList stopWords() const override
     {
         return QStringList() << "<SUF>" << "<PRE>" << "</PRE>" << "</SUF>" << "< EOT >" << "\\end"
@@ -35,27 +36,12 @@ public:
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        request["prompt"] = QString("<SUF>%1<PRE>%2<MID>")
-                                .arg(context.suffix.value_or(""), context.prefix.value_or(""));
-        request["system"] = context.systemPrompt.value_or("");
+        QString formattedPrompt = promptTemplate().arg(context.suffix, context.prefix);
+        request["prompt"] = formattedPrompt;
     }
     QString description() const override
     {
-        return "Specialized template for QML code completion with CodeLlama:\n\n"
-               "{\n"
-               "  \"prompt\": \"<SUF><code suffix><PRE><code prefix><MID>\",\n"
-               "  \"system\": \"<system prompt>\"\n"
-               "}\n\n"
-               "Specifically optimized for QML/JavaScript code completion.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: <SUF>%1<PRE>%2<MID>";
     }
 };
 

templates/GoogleAI.hpp

@@ -1,89 +0,0 @@
-/* 
- * Copyright (C) 2024 Petr Mironychev
- *
- * This file is part of QodeAssist.
- *
- * QodeAssist is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * QodeAssist is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#pragma once
-
-#include <QJsonArray>
-#include <QJsonObject>
-
-#include "llmcore/PromptTemplate.hpp"
-
-namespace QodeAssist::Templates {
-
-class GoogleAI : public LLMCore::PromptTemplate
-{
-public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-    QString name() const override { return "Google AI"; }
-    QStringList stopWords() const override { return QStringList(); }
-
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        QJsonArray contents;
-
-        if (context.systemPrompt && !context.systemPrompt->isEmpty()) {
-            request["system_instruction"] = QJsonObject{
-                {"parts", QJsonObject{{"text", context.systemPrompt.value()}}}};
-        }
-
-        for (const auto &msg : context.history.value()) {
-            QJsonObject content;
-            QJsonArray parts;
-
-            parts.append(QJsonObject{{"text", msg.content}});
-
-            QString role = msg.role;
-            if (role == "assistant") {
-                role = "model";
-            }
-
-            content["role"] = role;
-            content["parts"] = parts;
-            contents.append(content);
-        }
-
-        request["contents"] = contents;
-    }
-
-    QString description() const override
-    {
-        return "Template for Google AI models (Gemini):\n\n"
-               "{\n"
-               "  \"system_instruction\": {\"parts\": {\"text\": \"<system prompt>\"}},\n"
-               "  \"contents\": [\n"
-               "    {\n"
-               "      \"role\": \"user\",\n"
-               "      \"parts\": [{\"text\": \"<user message>\"}]\n"
-               "    },\n"
-               "    {\n"
-               "      \"role\": \"model\",\n"
-               "      \"parts\": [{\"text\": \"<assistant response>\"}]\n"
-               "    }\n"
-               "  ]\n"
-               "}\n\n"
-               "Supports proper role mapping, including model/user roles.";
-    }
-
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        return id == QodeAssist::LLMCore::ProviderID::GoogleAI;
-    }
-};
-
-} // namespace QodeAssist::Templates

templates/Llama2.hpp

@@ -29,58 +29,35 @@ class Llama2 : public LLMCore::PromptTemplate
 public:
     QString name() const override { return "Llama 2"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList() << "[INST]"; }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QJsonArray messages;
+        QJsonArray messages = request["messages"].toArray();
 
-        QString fullContent;
+        for (int i = 0; i < messages.size(); ++i) {
+            QJsonObject message = messages[i].toObject();
+            QString role = message["role"].toString();
+            QString content = message["content"].toString();
 
-        if (context.systemPrompt) {
-            fullContent
-                += QString("[INST]<<SYS>>\n%1\n<</SYS>>[/INST]\n").arg(context.systemPrompt.value());
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                if (msg.role == "user") {
-                    fullContent += QString("[INST]%1[/INST]\n").arg(msg.content);
-                } else if (msg.role == "assistant") {
-                    fullContent += msg.content + "\n";
-                }
+            QString formattedContent;
+            if (role == "system") {
+                formattedContent = QString("[INST]<<SYS>>\n%1\n<</SYS>>[/INST]\n").arg(content);
+            } else if (role == "user") {
+                formattedContent = QString("[INST]%1[/INST]\n").arg(content);
+            } else if (role == "assistant") {
+                formattedContent = content + "\n";
             }
-        }
 
-        messages.append(QJsonObject{{"role", "user"}, {"content", fullContent}});
+            message["content"] = formattedContent;
+            messages[i] = message;
+        }
 
         request["messages"] = messages;
     }
     QString description() const override
     {
-        return "Template for Llama 2 models:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\n"
-               "      \"role\": \"user\",\n"
-               "      \"content\": \"[INST]<<SYS>>\\n<system prompt>\\n<</SYS>>[/INST]\\n"
-               "<assistant response>\\n"
-               "[INST]<user message>[/INST]\\n\"\n"
-               "    }\n"
-               "  ]\n"
-               "}\n\n"
-               "Compatible with Ollama, LM Studio, and other services for Llama 2.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-        case QodeAssist::LLMCore::ProviderID::LMStudio:
-        case QodeAssist::LLMCore::ProviderID::OpenRouter:
-        case QodeAssist::LLMCore::ProviderID::OpenAICompatible:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: [INST]%1[/INST]\n";
     }
 };
 

templates/Llama3.hpp

@@ -30,64 +30,32 @@ class Llama3 : public LLMCore::PromptTemplate
 public:
     QString name() const override { return "Llama 3"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
+    QString promptTemplate() const override { return ""; }
     QStringList stopWords() const override
     {
         return QStringList() << "<|start_header_id|>" << "<|end_header_id|>" << "<|eot_id|>";
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QJsonArray messages;
+        QJsonArray messages = request["messages"].toArray();
 
-        if (context.systemPrompt) {
-            messages.append(QJsonObject{
-                {"role", "system"},
-                {"content",
-                 QString("<|start_header_id|>system<|end_header_id|>%2<|eot_id|>")
-                     .arg(context.systemPrompt.value())}});
-        }
+        for (int i = 0; i < messages.size(); ++i) {
+            QJsonObject message = messages[i].toObject();
+            QString role = message["role"].toString();
+            QString content = message["content"].toString();
 
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                messages.append(QJsonObject{
-                    {"role", msg.role},
-                    {"content",
-                     QString("<|start_header_id|>%1<|end_header_id|>%2<|eot_id|>")
-                         .arg(msg.role, msg.content)}});
-            }
+            message["content"]
+                = QString("<|start_header_id|>%1<|end_header_id|>%2<|eot_id|>").arg(role, content);
+
+            messages[i] = message;
         }
 
         request["messages"] = messages;
     }
     QString description() const override
     {
-        return "Template for Llama 3 models:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\n"
-               "      \"role\": \"system\",\n"
-               "      \"content\": \"<|start_header_id|>system<|end_header_id|><system "
-               "prompt><|eot_id|>\"\n"
-               "    },\n"
-               "    {\n"
-               "      \"role\": \"user\",\n"
-               "      \"content\": \"<|start_header_id|>user<|end_header_id|><user "
-               "message><|eot_id|>\"\n"
-               "    }\n"
-               "  ]\n"
-               "}\n\n"
-               "Compatible with Ollama, LM Studio, and OpenAI-compatible services for Llama 3.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-        case QodeAssist::LLMCore::ProviderID::LMStudio:
-        case QodeAssist::LLMCore::ProviderID::OpenRouter:
-        case QodeAssist::LLMCore::ProviderID::OpenAICompatible:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: "
+               "<|start_header_id|>%1<|end_header_id|>%2<|eot_id|>";
     }
 };
 

templates/MistralAI.hpp

@@ -1,106 +0,0 @@
-/* 
- * Copyright (C) 2024 Petr Mironychev
- *
- * This file is part of QodeAssist.
- *
- * QodeAssist is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * QodeAssist is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#pragma once
-
-#include <QJsonArray>
-
-#include "llmcore/PromptTemplate.hpp"
-
-namespace QodeAssist::Templates {
-
-class MistralAIFim : public LLMCore::PromptTemplate
-{
-public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::FIM; }
-    QString name() const override { return "Mistral AI FIM"; }
-    QStringList stopWords() const override { return QStringList(); }
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        request["prompt"] = context.prefix.value_or("");
-        request["suffix"] = context.suffix.value_or("");
-    }
-    QString description() const override
-    {
-        return "Template for MistralAI models with FIM support:\n\n"
-               "{\n"
-               "  \"prompt\": \"<code prefix>\",\n"
-               "  \"suffix\": \"<code suffix>\"\n"
-               "}\n\n"
-               "Optimized for code completion with MistralAI models.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::MistralAI:
-            return true;
-        default:
-            return false;
-        }
-    }
-};
-
-class MistralAIChat : public LLMCore::PromptTemplate
-{
-public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-    QString name() const override { return "Mistral AI Chat"; }
-    QStringList stopWords() const override { return QStringList(); }
-
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        QJsonArray messages;
-
-        if (context.systemPrompt) {
-            messages.append(
-                QJsonObject{{"role", "system"}, {"content", context.systemPrompt.value()}});
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
-            }
-        }
-
-        request["messages"] = messages;
-    }
-    QString description() const override
-    {
-        return "Template for MistralAI chat-capable models:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\"role\": \"system\", \"content\": \"<system prompt>\"},\n"
-               "    {\"role\": \"user\", \"content\": \"<user message>\"},\n"
-               "    {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
-               "  ]\n"
-               "}\n\n"
-               "Supports system messages and conversation history.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::MistralAI:
-            return true;
-        default:
-            return false;
-        }
-    }
-};
-
-} // namespace QodeAssist::Templates

templates/Ollama.hpp

@@ -25,84 +25,33 @@
 
 namespace QodeAssist::Templates {
 
-class OllamaFim : public LLMCore::PromptTemplate
+class OllamaAutoFim : public LLMCore::PromptTemplate
 {
 public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::FIM; }
-    QString name() const override { return "Ollama FIM"; }
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Fim; }
+    QString name() const override { return "Ollama Auto FIM"; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList(); }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        request["prompt"] = context.prefix.value_or("");
-        request["suffix"] = context.suffix.value_or("");
-        request["system"] = context.systemPrompt.value_or("");
-    }
-    QString description() const override
-    {
-        return "Default Ollama FIM (Fill-in-Middle) template with native format:\n\n"
-               "{\n"
-               "  \"prompt\": \"<code prefix>\",\n"
-               "  \"suffix\": \"<code suffix>\",\n"
-               "  \"system\": \"<system prompt>\"\n"
-               "}\n\n"
-               "Recommended for Ollama models with FIM capability.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-            return true;
-        default:
-            return false;
-        }
+        request["prompt"] = context.prefix;
+        request["suffix"] = context.suffix;
     }
+    QString description() const override { return "template will take from ollama modelfile"; }
 };
 
-class OllamaChat : public LLMCore::PromptTemplate
+class OllamaAutoChat : public LLMCore::PromptTemplate
 {
 public:
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-    QString name() const override { return "Ollama Chat"; }
+    QString name() const override { return "Ollama Auto Chat"; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList(); }
 
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QJsonArray messages;
-
-        if (context.systemPrompt) {
-            messages.append(
-                QJsonObject{{"role", "system"}, {"content", context.systemPrompt.value()}});
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
-            }
-        }
-
-        request["messages"] = messages;
-    }
-    QString description() const override
-    {
-        return "Template for Ollama Chat with message array format:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\"role\": \"system\", \"content\": \"<system prompt>\"},\n"
-               "    {\"role\": \"user\", \"content\": \"<user message>\"},\n"
-               "    {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
-               "  ]\n"
-               "}\n\n"
-               "Recommended for Ollama models with chat capability.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-            return true;
-        default:
-            return false;
-        }
     }
+    QString description() const override { return "template will take from ollama modelfile"; }
 };
 
 } // namespace QodeAssist::Templates

templates/OpenAI.hpp

@@ -30,45 +30,10 @@ class OpenAI : public LLMCore::PromptTemplate
 public:
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
     QString name() const override { return "OpenAI"; }
+    QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList(); }
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        QJsonArray messages;
-
-        if (context.systemPrompt) {
-            messages.append(
-                QJsonObject{{"role", "system"}, {"content", context.systemPrompt.value()}});
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
-            }
-        }
-
-        request["messages"] = messages;
-    }
-    QString description() const override
-    {
-        return "Template for OpenAI models (GPT series):\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\"role\": \"system\", \"content\": \"<system prompt>\"},\n"
-               "    {\"role\": \"user\", \"content\": \"<user message>\"},\n"
-               "    {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
-               "  ]\n"
-               "}\n\n"
-               "Standard Chat API format for OpenAI.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::OpenAI:
-            return true;
-        default:
-            return false;
-        }
-    }
+    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override {}
+    QString description() const override { return "OpenAI"; }
 };
 
 } // namespace QodeAssist::Templates

templates/OpenAICompatible.hpp

@@ -1,76 +0,0 @@
-/* 
- * Copyright (C) 2024 Petr Mironychev
- *
- * This file is part of QodeAssist.
- *
- * QodeAssist is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * QodeAssist is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#pragma once
-
-#include <QJsonArray>
-
-#include "llmcore/PromptTemplate.hpp"
-
-namespace QodeAssist::Templates {
-
-class OpenAICompatible : public LLMCore::PromptTemplate
-{
-public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-    QString name() const override { return "OpenAI Compatible"; }
-    QStringList stopWords() const override { return QStringList(); }
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        QJsonArray messages;
-
-        if (context.systemPrompt) {
-            messages.append(
-                QJsonObject{{"role", "system"}, {"content", context.systemPrompt.value()}});
-        }
-
-        if (context.history) {
-            for (const auto &msg : context.history.value()) {
-                messages.append(QJsonObject{{"role", msg.role}, {"content", msg.content}});
-            }
-        }
-
-        request["messages"] = messages;
-    }
-    QString description() const override
-    {
-        return "Generic template for OpenAI API-compatible services:\n\n"
-               "{\n"
-               "  \"messages\": [\n"
-               "    {\"role\": \"system\", \"content\": \"<system prompt>\"},\n"
-               "    {\"role\": \"user\", \"content\": \"<user message>\"},\n"
-               "    {\"role\": \"assistant\", \"content\": \"<assistant response>\"}\n"
-               "  ]\n"
-               "}\n\n"
-               "Works with any service implementing the OpenAI Chat API specification.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::OpenAICompatible:
-        case QodeAssist::LLMCore::ProviderID::OpenRouter:
-        case QodeAssist::LLMCore::ProviderID::LMStudio:
-            return true;
-        default:
-            return false;
-        }
-    }
-};
-
-} // namespace QodeAssist::Templates

templates/Qwen.hpp

@@ -28,32 +28,21 @@ class QwenFim : public LLMCore::PromptTemplate
 {
 public:
     QString name() const override { return "Qwen FIM"; }
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::FIM; }
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Fim; }
+    QString promptTemplate() const override
+    {
+        return "<|fim_prefix|>%1<|fim_suffix|>%2<|fim_middle|>";
+    }
     QStringList stopWords() const override { return QStringList() << "<|endoftext|>" << "<|EOT|>"; }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        request["prompt"] = QString("<|fim_prefix|>%1<|fim_suffix|>%2<|fim_middle|>")
-                                .arg(context.prefix.value_or(""), context.suffix.value_or(""));
-        request["system"] = context.systemPrompt.value_or("");
+        QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
+        request["prompt"] = formattedPrompt;
     }
     QString description() const override
     {
-        return "Template for Qwen models with FIM support:\n\n"
-               "{\n"
-               "  \"prompt\": \"<|fim_prefix|><code prefix><|fim_suffix|><code "
-               "suffix><|fim_middle|>\",\n"
-               "  \"system\": \"<system prompt>\"\n"
-               "}\n\n"
-               "Ideal for code completion with Qwen models.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: "
+               "<|fim_prefix|>%1<|fim_suffix|>%2<|fim_middle|>";
     }
 };
 

templates/StarCoder2Fim.hpp

@@ -26,8 +26,9 @@ namespace QodeAssist::Templates {
 class StarCoder2Fim : public LLMCore::PromptTemplate
 {
 public:
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::FIM; }
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Fim; }
     QString name() const override { return "StarCoder2 FIM"; }
+    QString promptTemplate() const override { return "<fim_prefix>%1<fim_suffix>%2<fim_middle>"; }
     QStringList stopWords() const override
     {
         return QStringList() << "<|endoftext|>" << "<file_sep>" << "<fim_prefix>" << "<fim_suffix>"
@@ -35,27 +36,13 @@ public:
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        request["prompt"] = QString("<fim_prefix>%1<fim_suffix>%2<fim_middle>")
-                                .arg(context.prefix.value_or(""), context.suffix.value_or(""));
-        request["system"] = context.systemPrompt.value_or("");
+        QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
+        request["prompt"] = formattedPrompt;
     }
     QString description() const override
     {
-        return "Template for StarCoder2 with FIM format:\n\n"
-               "{\n"
-               "  \"prompt\": \"<fim_prefix><code prefix><fim_suffix><code suffix><fim_middle>\",\n"
-               "  \"system\": \"<system prompt>\"\n"
-               "}\n\n"
-               "Includes stop words to prevent token duplication.";
-    }
-    bool isSupportProvider(LLMCore::ProviderID id) const override
-    {
-        switch (id) {
-        case QodeAssist::LLMCore::ProviderID::Ollama:
-            return true;
-        default:
-            return false;
-        }
+        return "The message will contain the following tokens: "
+               "<fim_prefix>%1<fim_suffix>%2<fim_middle>";
     }
 };
 

templates/Templates.hpp

@@ -21,19 +21,17 @@
 
 #include "llmcore/PromptTemplateManager.hpp"
 #include "templates/Alpaca.hpp"
+#include "templates/BasicChat.hpp"
 #include "templates/ChatML.hpp"
 #include "templates/Claude.hpp"
 #include "templates/CodeLlamaFim.hpp"
 #include "templates/CodeLlamaQMLFim.hpp"
-#include "templates/MistralAI.hpp"
-#include "templates/Ollama.hpp"
-#include "templates/OpenAI.hpp"
-#include "templates/OpenAICompatible.hpp"
-// #include "templates/CustomFimTemplate.hpp"
-// #include "templates/DeepSeekCoderFim.hpp"
-#include "templates/GoogleAI.hpp"
+#include "templates/CustomFimTemplate.hpp"
+#include "templates/DeepSeekCoderFim.hpp"
 #include "templates/Llama2.hpp"
 #include "templates/Llama3.hpp"
+#include "templates/Ollama.hpp"
+#include "templates/OpenAI.hpp"
 #include "templates/Qwen.hpp"
 #include "templates/StarCoder2Fim.hpp"
 
@@ -42,24 +40,21 @@ namespace QodeAssist::Templates {
 inline void registerTemplates()
 {
     auto &templateManager = LLMCore::PromptTemplateManager::instance();
-    templateManager.registerTemplate<OllamaChat>();
-    templateManager.registerTemplate<OllamaFim>();
     templateManager.registerTemplate<CodeLlamaFim>();
+    templateManager.registerTemplate<StarCoder2Fim>();
+    templateManager.registerTemplate<DeepSeekCoderFim>();
+    templateManager.registerTemplate<CustomTemplate>();
+    templateManager.registerTemplate<QwenFim>();
+    templateManager.registerTemplate<OllamaAutoFim>();
+    templateManager.registerTemplate<OllamaAutoChat>();
+    templateManager.registerTemplate<BasicChat>();
+    templateManager.registerTemplate<Llama3>();
+    templateManager.registerTemplate<ChatML>();
+    templateManager.registerTemplate<Alpaca>();
+    templateManager.registerTemplate<Llama2>();
     templateManager.registerTemplate<Claude>();
     templateManager.registerTemplate<OpenAI>();
-    templateManager.registerTemplate<MistralAIFim>();
-    templateManager.registerTemplate<MistralAIChat>();
     templateManager.registerTemplate<CodeLlamaQMLFim>();
-    templateManager.registerTemplate<ChatML>();
-    templateManager.registerTemplate<Llama2>();
-    templateManager.registerTemplate<Llama3>();
-    templateManager.registerTemplate<StarCoder2Fim>();
-    // templateManager.registerTemplate<DeepSeekCoderFim>();
-    // templateManager.registerTemplate<CustomTemplate>();
-    templateManager.registerTemplate<QwenFim>();
-    templateManager.registerTemplate<OpenAICompatible>();
-    templateManager.registerTemplate<Alpaca>();
-    templateManager.registerTemplate<GoogleAI>();
 }
 
 } // namespace QodeAssist::Templates