ZcAILib

Qt-based AI client library for chat completion APIs.
It currently provides:

• OpenAI / DeepSeek service switching
• model list fetching
• system prompt support
• non-streaming reply callbacks
• streaming reply callbacks based on SSE

Requirements

• Qt 6
• CMake 3.16+
• C++17

This project links against:

• Qt6::Core
• Qt6::Network
• Qt6::Widgets for the example app

Build

cmake -S . -B build
cmake --build build --config Debug

The example executable will be generated as:

• build/Debug/ZcAiLib-Example.exe on Windows with Visual Studio generators

Add To Your Project

If you use this repository directly, the simplest way is:

add_subdirectory(path/to/ZcAiLib)

target_link_libraries(YourApp PRIVATE
    ZcAiLib
    Qt6::Widgets
)

Then include:

#include "AiProvider.h"

Basic Usage

AiProvider *ai = new AiProvider(this);

ai->setServiceType(AiProvider::DeepSeek);
ai->setApiKey("YOUR_API_KEY");
ai->setModel("deepseek-chat");
ai->setSystemPrompt("You are a helpful assistant.");

connect(ai, &AiProvider::replyReceived, this, [](const QString &reply) {
    qDebug() << "AI reply:" << reply;
});

connect(ai, &AiProvider::errorOccurred, this, [](const QString &error) {
    qWarning() << "AI error:" << error;
});

ai->chat("Hello");

Streaming Usage

Streaming is disabled by default. Enable it explicitly before calling chat().

AiProvider *ai = new AiProvider(this);

ai->setServiceType(AiProvider::OpenAI);
ai->setApiKey("YOUR_API_KEY");
ai->setModel("gpt-3.5-turbo");
ai->setStreamEnabled(true);

connect(ai, &AiProvider::replyChunkReceived, this, [](const QString &chunk) {
    qDebug().noquote() << chunk;
});

connect(ai, &AiProvider::replyReceived, this, [](const QString &fullReply) {
    qDebug() << "stream finished:" << fullReply;
});

connect(ai, &AiProvider::errorOccurred, this, [](const QString &error) {
    qWarning() << error;
});

ai->chat("Write a short poem about Qt.");

Streaming behavior

• replyChunkReceived(...) is emitted for every parsed content chunk
• replyReceived(...) is emitted once after the full streamed reply is complete
• errorOccurred(...) is emitted on HTTP or API errors

This means existing code that only listens to replyReceived(...) still works after streaming is added.

Fetch Available Models

connect(ai, &AiProvider::modelsReceived, this, [](const QList<AiProvider::ModelInfo> &models) {
    for (const auto &model : models) {
        qDebug() << model.id << model.ownedBy << model.created;
    }
});

ai->fetchModels();

ModelInfo contains:

• id
• created
• ownedBy
• permissions

Service Configuration

Built-in service types:

• AiProvider::OpenAI
• AiProvider::DeepSeek
• AiProvider::Custom

For custom services:

ai->setServiceType(AiProvider::Custom);
ai->setBaseUrl("https://your-domain/v1");
ai->setApiKey("YOUR_API_KEY");
ai->setModel("your-model-id");

If you need to point at a fully custom chat endpoint:

ai->setServiceType(AiProvider::Custom);
ai->setApiUrl("https://your-domain/v1/chat/completions");
ai->setApiKey("YOUR_API_KEY");
ai->setModel("your-model-id");

Public API Summary

Main methods:

• setServiceType(AiProvider::ServiceType type)
• setApiKey(const QString &apiKey)
• setApiUrl(const QString &url)
• setBaseUrl(const QString &baseUrl)
• setModel(const QString &model)
• setStreamEnabled(bool enabled)
• setSystemPrompt(const QString &prompt)
• fetchModels()
• chat(const QString &message)

Main signals:

• replyReceived(const QString &reply)
• replyChunkReceived(const QString &chunk)
• modelsReceived(const QList<ModelInfo> &models)
• errorOccurred(const QString &error)

Example

The example UI is in example/mainwindow.cpp.

It demonstrates:

• switching between OpenAI and DeepSeek
• fetching model lists from the API
• setting a system prompt
• enabling/disabling stream output
• rendering streamed output incrementally in a QTextEdit

Notes

• Streaming support depends on the upstream API supporting SSE chat completion responses.
• The library currently targets chat-completions style APIs.
• replyReceived(...) remains the final completion signal for both streaming and non-streaming requests.