A native macOS / windows / Linux database IDE built with Swift and AppKit. Connect to PostgreSQL, MySQL, SQLite, and Redis from a single app with a fast, keyboard-driven interface.
C++
#pragma once
#include <string>
#include <vector>
#include <functional>
#include <map>
namespace DBModels
{
// Index order is persisted in settings (aiProviderIndex) and bound to
// the SettingsPage ComboBox. NEVER reorder existing entries — only
// append new providers at the end.
enum class AiProvider
{
Anthropic = 0,
OpenAI = 1,
Ollama = 2,
Gemini = 3,
OpenRouter = 4,
ChatGPT = 5, // Subscription-based via OAuth (no API key)
Requesty = 6, // OpenAI-compatible router (https://router.requesty.ai/v1)
Opper = 7, // OpenAI-compatible gateway (https://api.opper.ai/v3/compat)
};
struct AiConfig
{
AiProvider provider = AiProvider::Anthropic;
std::wstring apiKey;
std::wstring model; // e.g. "claude-sonnet-4-20250514", "gpt-4o", "llama3", "gemini-2.0-flash"
std::wstring ollamaEndpoint; // e.g. "http://localhost:11434"
};
struct ChatMessage
{
std::wstring role; // "user", "assistant", "system"
std::wstring content;
};
// Result of a FetchModels() call. success=false → errorMessage holds
// a user-visible reason (network error, bad key, unsupported
// endpoint); models may still be empty on success for providers
// without any models configured.
struct ModelListResult
{
bool success = false;
std::vector<std::wstring> models;
std::wstring errorMessage;
};
// AI service for text-to-SQL and chat interactions
class AiService
{
public:
void SetConfig(const AiConfig& config) { config_ = config; }
const AiConfig& GetConfig() const { return config_; }
// Send chat completion request (blocking)
// Returns assistant response
std::wstring SendChat(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt = L"");
// Convenience: text-to-SQL with schema context
std::wstring TextToSql(
const std::wstring& naturalLanguage,
const std::wstring& schemaDescription);
// List available models for the given provider config. BLOCKS on
// the provider's HTTP models endpoint — callers on the UI thread
// must dispatch to a worker. Used by Settings to populate an
// editable ComboBox so users pick from a live list instead of
// memorizing model IDs, while still allowing a free-typed name
// for custom OpenAI-compatible domains.
static ModelListResult FetchModels(const AiConfig& config);
private:
AiConfig config_;
// Provider-specific HTTP calls
std::wstring CallAnthropic(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallOpenAI(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallOllama(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallGemini(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallOpenRouter(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallRequesty(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallOpper(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
std::wstring CallChatGPT(
const std::vector<ChatMessage>& messages,
const std::wstring& systemPrompt);
// UTF-8 helpers
static std::string toUtf8(const std::wstring& wstr);
static std::wstring fromUtf8(const std::string& str);
};
}