llm_interface.h

#pragma once
 
#include <agents-cpp/types.h>
#include <agents-cpp/tool.h>
#include <agents-cpp/coroutine_utils.h>
#include <functional>
#include <memory>
#include <optional>
#include <vector>
 
namespace agents {

struct LLMOptions {
    double temperature = 0.7;
    int max_tokens = 1024;
    double top_p = 1.0;
    double presence_penalty = 0.0;
    double frequency_penalty = 0.0;
    int timeout_ms = 30000; // 30 seconds
    std::vector<String> stop_sequences;
    std::optional<JsonObject> response_schema;
    std::optional<String> response_mime_type;
};

class LLMInterface {
public:
    virtual ~LLMInterface() = default;

    virtual std::vector<String> getAvailableModels() = 0;

    virtual void setModel(const String& model) = 0;

    virtual String getModel() const = 0;

    virtual void setApiKey(const String& api_key) = 0;

    virtual void setApiBase(const String& api_base) = 0;

    virtual void setOptions(const LLMOptions& options) = 0;

    virtual LLMOptions getOptions() const = 0;

    virtual LLMResponse complete(const String& prompt);

    virtual LLMResponse complete(const std::vector<Message>& messages);

    virtual LLMResponse completeWithTools(
        const std::vector<Message>& messages,
        const std::vector<JsonObject>& tools_schema
    );

    virtual LLMResponse chat(const std::vector<Message>& messages) = 0;

    virtual LLMResponse chatWithTools(
        const std::vector<Message>& messages,
        const std::vector<std::shared_ptr<Tool>>& tools
    ) = 0;

    virtual void streamChat(
        const std::vector<Message>& messages,
        std::function<void(const String&, bool)> callback
    ) = 0;

    virtual Task<LLMResponse> completeAsync(const String& prompt);

    virtual Task<LLMResponse> completeAsync(const std::vector<Message>& messages);

    virtual Task<LLMResponse> chatAsync(const std::vector<Message>& messages);

    virtual Task<LLMResponse> chatWithToolsAsync(
        const std::vector<Message>& messages,
        const std::vector<std::shared_ptr<Tool>>& tools
    );

    virtual AsyncGenerator<String> streamChatAsync(
        const std::vector<Message>& messages
    );
};

std::shared_ptr<LLMInterface> createLLM(
    const String& provider,
    const String& api_key,
    const String& model = ""
);
 
} // namespace agents