feat: Add DeepSeek provider

CMakeLists.txt

@@ -118,6 +118,7 @@ add_qtc_plugin(QodeAssist
     providers/OpenAIResponsesProvider.hpp providers/OpenAIResponsesProvider.cpp
     providers/QwenProvider.hpp providers/QwenProvider.cpp
     providers/QwenResponsesProvider.hpp providers/QwenResponsesProvider.cpp
+    providers/DeepSeekProvider.hpp providers/DeepSeekProvider.cpp
     QodeAssist.qrc
     LSPCompletion.hpp
     LLMSuggestion.hpp LLMSuggestion.cpp

README.md

@@ -6,7 +6,7 @@
 [![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg)](https://www.gnu.org/licenses/gpl-3.0)
 [![Discord](https://dcbadge.limes.pink/api/server/BGMkUsXUgf?style=flat)](https://discord.gg/BGMkUsXUgf)
 
-![qodeassist-icon](https://github.com/user-attachments/assets/dc336712-83cb-440d-8761-8d0a31de898d) **QodeAssist** brings a full AI coding workflow to Qt Creator for C++ and QML — smart code completion, multi-panel chat, inline quick refactoring, and project-aware tool calling. It works with local runtimes (Ollama, llama.cpp, LM Studio) and cloud providers (Claude, OpenAI, Google AI, Mistral, Qwen), can run as an **MCP server** so other clients reuse its project context, and can also act as an **MCP client** to consume tools from external MCP servers (authenticated MCP servers are not supported yet).
+![qodeassist-icon](https://github.com/user-attachments/assets/dc336712-83cb-440d-8761-8d0a31de898d) **QodeAssist** brings a full AI coding workflow to Qt Creator for C++ and QML — smart code completion, multi-panel chat, inline quick refactoring, and project-aware tool calling. It works with local runtimes (Ollama, llama.cpp, LM Studio) and cloud providers (Claude, OpenAI, Google AI, Mistral, Qwen, DeepSeek), can run as an **MCP server** so other clients reuse its project context, and can also act as an **MCP client** to consume tools from external MCP servers (authenticated MCP servers are not supported yet).
 
 ⚠️ **Important Notice About Paid Providers**
 > When using paid providers like Claude, OpenRouter or OpenAI-compatible services:
@@ -39,7 +39,8 @@ QodeAssist enhances Qt Creator with AI-powered coding assistance:
 - **MCP Server** — expose QodeAssist's project-aware tools to external MCP clients (Claude Code, VS Code, Claude Desktop via bridge)
 - **MCP Client Hub** — connect QodeAssist to external MCP servers and use their tools in Chat and Quick Refactor (authenticated MCP servers are not supported yet)
 - **File Context** — attach, link, or auto-sync open editor files for richer prompts
-- **Many Providers** — Ollama, llama.cpp, LM Studio (Chat + Responses), Claude, OpenAI (Chat + Responses), Google AI, Mistral, Codestral, OpenRouter, Qwen (OpenAI + Responses), any OpenAI-compatible endpoint
+- **Many Providers** — Ollama, llama.cpp, LM Studio (Chat + Responses), Claude, OpenAI (Chat + Responses), Google AI, Mistral, Codestral, OpenRouter, Qwen (OpenAI + Responses), DeepSeek, any OpenAI-compatible endpoint
+- **Reasoning / Thinking** — streamed chain-of-thought is shown for reasoning models across Claude, Google, OpenAI Responses, and any OpenAI-compatible endpoint that returns `reasoning_content` (DeepSeek, Qwen QwQ/Qwen3-Thinking, LM Studio, OpenRouter, …)
 - **Customizable** — per-project rules (`.qodeassist/rules/`), agent roles, reusable refactor templates, full prompt-template control
 
 **Join our [Discord Community](https://discord.gg/BGMkUsXUgf)** to get support and connect with other users!
@@ -170,6 +171,7 @@ The Quick Setup feature provides one-click configuration for popular cloud AI mo
    - **Mistral AI** (Codestral 2501)
    - **Google AI** (Gemini 2.5 Flash)
    - **Qwen** (Qwen3.6 Plus, Qwen3.7 Max)
+   - **DeepSeek** (DeepSeek V4 Flash, DeepSeek V4 Pro)
 3. **Configure API Key** - Click "Configure API Key" button and enter your API key in Provider Settings
 
 All settings (provider, model, template, URL) are configured automatically. Just add your API key and you're ready to go!
@@ -191,6 +193,7 @@ For advanced users or local models, choose your preferred provider and follow th
 - **[Mistral AI](docs/mistral-configuration.md)** / **Codestral**
 - **[Google AI](docs/google-ai-configuration.md)** — Gemini
 - **Qwen (Alibaba)** — DashScope OpenAI-compatible Chat and Responses endpoints
+- **DeepSeek** — `deepseek-chat` and `deepseek-reasoner` (reasoning shown as thinking)
 - **OpenAI-compatible** — OpenRouter and any custom endpoint
 
 ### Recommended Models for Best Experience
@@ -255,7 +258,7 @@ Configure in: `Tools → Options → QodeAssist → Code Completion → General
 - **[Chat Summarization](docs/chat-summarization.md)** - Compress long conversations into AI-generated summaries
 - **[File Context](docs/file-context.md)** - Attach or link files for better context
 - Automatic syncing with open editor files (optional)
-- Extended thinking mode (Claude, other providers in plan) - Enable deeper reasoning for complex tasks
+- Extended thinking / reasoning mode - shows streamed chain-of-thought for reasoning models (Claude, Google, OpenAI Responses, and OpenAI-compatible endpoints returning `reasoning_content` such as DeepSeek, Qwen, LM Studio, OpenRouter)
 
 ### Quick Refactoring
 - Inline code refactoring directly in the editor with AI assistance

pluginllmcore/ProviderID.hpp

@@ -15,6 +15,7 @@ enum class ProviderID {
     OpenRouter,
     GoogleAI,
     LlamaCpp,
-    Qwen
+    Qwen,
+    DeepSeek
 };
 }

providers/DeepSeekProvider.cpp

@@ -0,0 +1,110 @@
+// Copyright (C) 2024-2026 Petr Mironychev
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include "DeepSeekProvider.hpp"
+
+#include <LLMQore/ToolsManager.hpp>
+#include "logger/Logger.hpp"
+#include "settings/ChatAssistantSettings.hpp"
+#include "settings/CodeCompletionSettings.hpp"
+#include "settings/GeneralSettings.hpp"
+#include "settings/ProviderSettings.hpp"
+#include "settings/QuickRefactorSettings.hpp"
+#include "tools/ToolsRegistration.hpp"
+
+#include <QJsonArray>
+#include <QJsonDocument>
+#include <QJsonObject>
+
+namespace QodeAssist::Providers {
+
+DeepSeekProvider::DeepSeekProvider(QObject *parent)
+    : PluginLLMCore::Provider(parent)
+    , m_client(new ::LLMQore::OpenAIClient(QString(), QString(), QString(), this))
+{
+    Tools::registerQodeAssistTools(m_client->tools());
+}
+
+QString DeepSeekProvider::name() const
+{
+    return "DeepSeek";
+}
+
+QString DeepSeekProvider::apiKey() const
+{
+    return Settings::providerSettings().deepSeekApiKey();
+}
+
+QString DeepSeekProvider::url() const
+{
+    return "https://api.deepseek.com";
+}
+
+QFuture<QList<QString>> DeepSeekProvider::getInstalledModels(const QString &url)
+{
+    m_client->setUrl(url);
+    m_client->setApiKey(apiKey());
+    return m_client->listModels();
+}
+
+PluginLLMCore::ProviderID DeepSeekProvider::providerID() const
+{
+    return PluginLLMCore::ProviderID::DeepSeek;
+}
+
+PluginLLMCore::ProviderCapabilities DeepSeekProvider::capabilities() const
+{
+    return PluginLLMCore::ProviderCapability::Tools
+           | PluginLLMCore::ProviderCapability::Thinking
+           | PluginLLMCore::ProviderCapability::ModelListing;
+}
+
+void DeepSeekProvider::prepareRequest(
+    QJsonObject &request,
+    PluginLLMCore::PromptTemplate *prompt,
+    PluginLLMCore::ContextData context,
+    PluginLLMCore::RequestType type,
+    bool isToolsEnabled,
+    bool isThinkingEnabled)
+{
+    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.useFrequencyPenalty())
+            request["frequency_penalty"] = settings.frequencyPenalty();
+        if (settings.usePresencePenalty())
+            request["presence_penalty"] = settings.presencePenalty();
+    };
+
+    if (type == PluginLLMCore::RequestType::CodeCompletion) {
+        applyModelParams(Settings::codeCompletionSettings());
+    } else if (type == PluginLLMCore::RequestType::QuickRefactoring) {
+        applyModelParams(Settings::quickRefactorSettings());
+    } else {
+        applyModelParams(Settings::chatAssistantSettings());
+    }
+
+    if (isToolsEnabled) {
+        auto toolsDefinitions = m_client->tools()->getToolsDefinitions();
+        if (!toolsDefinitions.isEmpty()) {
+            request["tools"] = toolsDefinitions;
+            LOG_MESSAGE(QString("Added %1 tools to DeepSeek request").arg(toolsDefinitions.size()));
+        }
+    }
+}
+
+::LLMQore::BaseClient *DeepSeekProvider::client() const
+{
+    return m_client;
+}
+
+} // namespace QodeAssist::Providers

providers/DeepSeekProvider.hpp

@@ -0,0 +1,37 @@
+// Copyright (C) 2024-2026 Petr Mironychev
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <LLMQore/OpenAIClient.hpp>
+#include <pluginllmcore/Provider.hpp>
+
+namespace QodeAssist::Providers {
+
+class DeepSeekProvider : public PluginLLMCore::Provider
+{
+    Q_OBJECT
+public:
+    explicit DeepSeekProvider(QObject *parent = nullptr);
+
+    QString name() const override;
+    QString url() const override;
+    void prepareRequest(
+        QJsonObject &request,
+        PluginLLMCore::PromptTemplate *prompt,
+        PluginLLMCore::ContextData context,
+        PluginLLMCore::RequestType type,
+        bool isToolsEnabled,
+        bool isThinkingEnabled) override;
+    QFuture<QList<QString>> getInstalledModels(const QString &url) override;
+    PluginLLMCore::ProviderID providerID() const override;
+    PluginLLMCore::ProviderCapabilities capabilities() const override;
+
+    ::LLMQore::BaseClient *client() const override;
+    QString apiKey() const override;
+
+private:
+    ::LLMQore::OpenAIClient *m_client;
+};
+
+} // namespace QodeAssist::Providers

providers/LMStudioProvider.cpp

@@ -57,7 +57,8 @@ PluginLLMCore::ProviderID LMStudioProvider::providerID() const
 PluginLLMCore::ProviderCapabilities LMStudioProvider::capabilities() const
 {
     return PluginLLMCore::ProviderCapability::Tools | PluginLLMCore::ProviderCapability::Image
-           | PluginLLMCore::ProviderCapability::ModelListing;
+           | PluginLLMCore::ProviderCapability::ModelListing
+           | PluginLLMCore::ProviderCapability::Thinking;
 }
 
 void LMStudioProvider::prepareRequest(

providers/MistralAIProvider.cpp

@@ -55,7 +55,8 @@ PluginLLMCore::ProviderID MistralAIProvider::providerID() const
 PluginLLMCore::ProviderCapabilities MistralAIProvider::capabilities() const
 {
     return PluginLLMCore::ProviderCapability::Tools | PluginLLMCore::ProviderCapability::Image
-           | PluginLLMCore::ProviderCapability::ModelListing;
+           | PluginLLMCore::ProviderCapability::ModelListing
+           | PluginLLMCore::ProviderCapability::Thinking;
 }
 
 void MistralAIProvider::prepareRequest(

providers/OllamaCompatProvider.cpp

@@ -113,7 +113,8 @@ PluginLLMCore::ProviderID OllamaCompatProvider::providerID() const
 PluginLLMCore::ProviderCapabilities OllamaCompatProvider::capabilities() const
 {
     return PluginLLMCore::ProviderCapability::Tools | PluginLLMCore::ProviderCapability::Image
-           | PluginLLMCore::ProviderCapability::ModelListing;
+           | PluginLLMCore::ProviderCapability::ModelListing
+           | PluginLLMCore::ProviderCapability::Thinking;
 }
 
 ::LLMQore::BaseClient *OllamaCompatProvider::client() const

providers/OpenAICompatProvider.cpp

@@ -98,7 +98,8 @@ PluginLLMCore::ProviderID OpenAICompatProvider::providerID() const
 
 PluginLLMCore::ProviderCapabilities OpenAICompatProvider::capabilities() const
 {
-    return PluginLLMCore::ProviderCapability::Tools | PluginLLMCore::ProviderCapability::Image;
+    return PluginLLMCore::ProviderCapability::Tools | PluginLLMCore::ProviderCapability::Image
+           | PluginLLMCore::ProviderCapability::Thinking;
 }
 
 ::LLMQore::BaseClient *OpenAICompatProvider::client() const

providers/OpenAIProvider.cpp

@@ -129,7 +129,8 @@ PluginLLMCore::ProviderID OpenAIProvider::providerID() const
 PluginLLMCore::ProviderCapabilities OpenAIProvider::capabilities() const
 {
     return PluginLLMCore::ProviderCapability::Tools | PluginLLMCore::ProviderCapability::Image
-           | PluginLLMCore::ProviderCapability::ModelListing;
+           | PluginLLMCore::ProviderCapability::ModelListing
+           | PluginLLMCore::ProviderCapability::Thinking;
 }
 
 ::LLMQore::BaseClient *OpenAIProvider::client() const

providers/Providers.hpp

@@ -5,6 +5,7 @@
 
 #include "pluginllmcore/ProvidersManager.hpp"
 #include "providers/ClaudeProvider.hpp"
+#include "providers/DeepSeekProvider.hpp"
 #include "providers/CodestralProvider.hpp"
 #include "providers/GoogleAIProvider.hpp"
 #include "providers/LMStudioProvider.hpp"
@@ -40,6 +41,7 @@ inline void registerProviders()
     providerManager.registerProvider<CodestralProvider>();
     providerManager.registerProvider<QwenProvider>();
     providerManager.registerProvider<QwenResponsesProvider>();
+    providerManager.registerProvider<DeepSeekProvider>();
 }
 
 } // namespace QodeAssist::Providers

settings/ConfigurationManager.cpp

@@ -121,6 +121,28 @@ QVector<AIConfiguration> ConfigurationManager::getPredefinedConfigurations(
     qwenMax.type = type;
     qwenMax.isPredefined = true;
 
+    AIConfiguration deepSeekFlash;
+    deepSeekFlash.id = "preset_deepseek_flash";
+    deepSeekFlash.name = "DeepSeek V4 Flash";
+    deepSeekFlash.provider = "DeepSeek";
+    deepSeekFlash.model = "deepseek-v4-flash";
+    deepSeekFlash.url = "https://api.deepseek.com";
+    deepSeekFlash.customEndpoint = "";
+    deepSeekFlash.templateName = "OpenAI Compatible";
+    deepSeekFlash.type = type;
+    deepSeekFlash.isPredefined = true;
+
+    AIConfiguration deepSeekPro;
+    deepSeekPro.id = "preset_deepseek_pro";
+    deepSeekPro.name = "DeepSeek V4 Pro";
+    deepSeekPro.provider = "DeepSeek";
+    deepSeekPro.model = "deepseek-v4-pro";
+    deepSeekPro.url = "https://api.deepseek.com";
+    deepSeekPro.customEndpoint = "";
+    deepSeekPro.templateName = "OpenAI Compatible";
+    deepSeekPro.type = type;
+    deepSeekPro.isPredefined = true;
+
     AIConfiguration gpt;
     gpt.id = "preset_gpt";
     gpt.name = "gpt-5.5";
@@ -141,6 +163,8 @@ QVector<AIConfiguration> ConfigurationManager::getPredefinedConfigurations(
     presets.append(geminiFlash);
     presets.append(qwenPlus);
     presets.append(qwenMax);
+    presets.append(deepSeekFlash);
+    presets.append(deepSeekPro);
 
     return presets;
 }

settings/ProviderSettings.cpp

@@ -132,6 +132,15 @@ ProviderSettings::ProviderSettings()
     qwenApiKey.setDefaultValue("");
     qwenApiKey.setAutoApply(true);
 
+    // DeepSeek Settings
+    deepSeekApiKey.setSettingsKey(Constants::DEEPSEEK_API_KEY);
+    deepSeekApiKey.setLabelText(Tr::tr("DeepSeek API Key:"));
+    deepSeekApiKey.setDisplayStyle(Utils::StringAspect::LineEditDisplay);
+    deepSeekApiKey.setPlaceHolderText(Tr::tr("Enter your API key here"));
+    deepSeekApiKey.setHistoryCompleter(Constants::DEEPSEEK_API_KEY_HISTORY);
+    deepSeekApiKey.setDefaultValue("");
+    deepSeekApiKey.setAutoApply(true);
+
     resetToDefaults.m_buttonText = Tr::tr("Reset Page to Defaults");
 
     readSettings();
@@ -163,6 +172,8 @@ ProviderSettings::ProviderSettings()
             Group{title(Tr::tr("llama.cpp Settings")), Column{llamaCppApiKey}},
             Space{8},
             Group{title(Tr::tr("Qwen (Alibaba) Settings")), Column{qwenApiKey}},
+            Space{8},
+            Group{title(Tr::tr("DeepSeek Settings")), Column{deepSeekApiKey}},
             Stretch{1}};
     });
 }
@@ -201,6 +212,9 @@ void ProviderSettings::setupConnections()
         llamaCppApiKey.writeSettings();
     });
     connect(&qwenApiKey, &ButtonAspect::changed, this, [this]() { qwenApiKey.writeSettings(); });
+    connect(&deepSeekApiKey, &ButtonAspect::changed, this, [this]() {
+        deepSeekApiKey.writeSettings();
+    });
 }
 
 void ProviderSettings::resetSettingsToDefaults()
@@ -224,6 +238,7 @@ void ProviderSettings::resetSettingsToDefaults()
         resetAspect(ollamaBasicAuthApiKey);
         resetAspect(llamaCppApiKey);
         resetAspect(qwenApiKey);
+        resetAspect(deepSeekApiKey);
         writeSettings();
     }
 }

settings/ProviderSettings.hpp

@@ -29,6 +29,7 @@ public:
     Utils::StringAspect ollamaBasicAuthApiKey{this};
     Utils::StringAspect llamaCppApiKey{this};
     Utils::StringAspect qwenApiKey{this};
+    Utils::StringAspect deepSeekApiKey{this};
 
 private:
     void setupConnections();

settings/SettingsConstants.hpp

@@ -165,6 +165,8 @@ const char LLAMA_CPP_API_KEY[] = "QodeAssist.llamaCppApiKey";
 const char LLAMA_CPP_API_KEY_HISTORY[] = "QodeAssist.llamaCppApiKeyHistory";
 const char QWEN_API_KEY[] = "QodeAssist.qwenApiKey";
 const char QWEN_API_KEY_HISTORY[] = "QodeAssist.qwenApiKeyHistory";
+const char DEEPSEEK_API_KEY[] = "QodeAssist.deepSeekApiKey";
+const char DEEPSEEK_API_KEY_HISTORY[] = "QodeAssist.deepSeekApiKeyHistory";
 
 const char CLAUDE_ENABLE_PROMPT_CACHING[] = "QodeAssist.claudeEnablePromptCaching";
 const char CLAUDE_USE_EXTENDED_CACHE_TTL[] = "QodeAssist.claudeUseExtendedCacheTTL";

templates/OpenAICompatible.hpp

@@ -82,6 +82,7 @@ public:
         case PluginLLMCore::ProviderID::LMStudio:
         case PluginLLMCore::ProviderID::LlamaCpp:
         case PluginLLMCore::ProviderID::Qwen:
+        case PluginLLMCore::ProviderID::DeepSeek:
             return true;
         default:
             return false;