autonomous_agent_example.cpp

Autonomous Agent Example.

Autonomous Agent Example

Version

0.1

Date

2025-07-20

Copyright

Copyright (c) 2025 Edge AI, LLC. All rights reserved.

#include <agents-cpp/agents/autonomous_agent.h>
#include <agents-cpp/config_loader.h>
#include <agents-cpp/logger.h>
#include <agents-cpp/tools/tool_registry.h>
 
#include <chrono>
#include <iostream>
 
using namespace agents;
 
// Custom callback to print detailed agent steps
void detailedStepCallback(const AutonomousAgent::Step& step) {
    Logger::info("\n=== STEP ===");
    Logger::info("Description: {}", step.description);
    Logger::info("Status: {}", step.status);
 
    if (step.success) {
        Logger::info("\nResult: {}", step.result.dump(2));
    } else {
        Logger::error("\nFailed!");
    }
 
    Logger::info("\n------------------------------------");
}
 
// Custom callback for human-in-the-loop
bool detailedHumanApproval(const String& message, const JsonObject& context, String& modifications) {
    if (!context.empty()) {
        Logger::info("\nContext Information:");
        Logger::info("{}", context.dump(2));
    }
 
    Logger::info("\n🔔 HUMAN APPROVAL REQUIRED 🔔");
    Logger::info("{}", message);
    Logger::info("\nApprove this step? (y/n/m - y: approve, n: reject, m: modify): ");
    char response;
    std::cin >> response;
    std::cin.ignore(); // Clear the newline
 
    if (response == 'm' || response == 'M') {
        Logger::info("Enter your modifications or instructions: ");
        String user_modifications;
        std::getline(std::cin, user_modifications);
 
        // Set modifications output parameter
        modifications = user_modifications;
 
        // Return true to continue with modifications
        Logger::info("Continuing with your modifications...");
        return true;
    }
 
    return (response == 'y' || response == 'Y');
}
 
int main() {
    // Initialize the logger
    Logger::init(Logger::Level::INFO);
 
    // Get model choice from user
    Logger::info("Select LLM provider (1 for OpenAI, 2 for Anthropic, 3 for Google): ");
    int provider_choice;
    std::cin >> provider_choice;
    std::cin.ignore(); // Clear the newline
 
    // Create LLM based on user choice
    std::shared_ptr<LLMInterface> llm;
 
    // Check if we have any API keys configured
    auto& config = ConfigLoader::getInstance();
 
    // Proceed based on provider choice and API key in the environment
    if (provider_choice == 1) {
        bool hasOpenAI = config.has("OPENAI_API_KEY");
        if (hasOpenAI) {
            llm = createLLM("openai", config.get("OPENAI_API_KEY"), "gpt-4o-2024-05-13");
        } else {
            Logger::error("OPENAI_API_KEY not found in environment.");
            return EXIT_FAILURE;
        }
    } else if (provider_choice == 2) {
        // Note: This assumes the user has the Anthropic API key in the environment
        bool hasAnthropic = config.has("ANTHROPIC_API_KEY");
        if (hasAnthropic) {
            llm = createLLM("anthropic", config.get("ANTHROPIC_API_KEY"), "claude-3-5-sonnet-20240620");
        } else {
            Logger::error("ANTHROPIC_API_KEY not found in environment.");
            return EXIT_FAILURE;
        }
    } else if (provider_choice == 3) {
        bool hasGoogle = config.has("GEMINI_API_KEY");
        if (hasGoogle) {
            llm = createLLM("google", config.get("GEMINI_API_KEY"), "gemini-2.0-flash");
        } else {
            Logger::error("GEMINI_API_KEY not found in environment.");
            return EXIT_FAILURE;
        }
    } else {
        Logger::error("Invalid provider choice.");
        return EXIT_FAILURE;
    }
 
    // Configure LLM options
    LLMOptions options;
    options.temperature = 0.2;
    options.max_tokens = 4096;
    llm->setOptions(options);
 
    // Create agent context
    auto context = std::make_shared<AgentContext>();
    context->setLLM(llm);
 
    // Set system prompt for the context
    context->setSystemPrompt(
        "You are a helpful, autonomous assistant with access to tools."
        "You can use these tools to accomplish tasks for the user."
        "Think step by step and be thorough in your approach."
    );
 
    // Register tools
    context->registerTool(tools::createWebSearchTool());
    context->registerTool(tools::createWikipediaTool());
    // Uncomment when tools are implemented
    // context->registerTool(tools::createWeatherTool());
    // context->registerTool(tools::createCalculatorTool());
 
    // Create a custom tool
    auto summarize_tool = createTool(
        "summarize",
        "Summarizes a long piece of text into a concise summary",
        {
            {"text", "The text to summarize", "string", true},
            {"max_length", "Maximum length of summary in words", "integer", false}
        },
        [context](const JsonObject& params) -> ToolResult {
            String text = params["text"];
            int max_length = params.contains("max_length") ? params["max_length"].get<int>() : 100;
 
            // Create a specific context for summarization
            auto summary_context = std::make_shared<AgentContext>(*context);
            summary_context->setSystemPrompt(
                "You are a summarization assistant. Your task is to create concise, accurate summaries "
                "that capture the main points of the provided text."
            );
 
            String prompt = "Summarize the following text in no more than " +
                            std::to_string(max_length) + " words:\n\n" + text;
 
            LLMResponse llm_response = summary_context->getLLM()->complete(prompt);
            String summary = llm_response.content;
 
            return ToolResult{
                true,
                summary,
                {{"summary", prompt + "\n\n" + summary}}
            };
        }
    );
    context->registerTool(summarize_tool);
 
    // Allow the user to choose planning strategy
    Logger::info("Select planning strategy:");
    Logger::info("1. ReAct (Open-ended, evolving task)");
    Logger::info("2. Plan-and-Execute (Complex, structured task)");
    Logger::info("Choice: ");
    int strategy_choice;
    std::cin >> strategy_choice;
    std::cin.ignore(); // Clear the newline
 
    // Create the agent
    AutonomousAgent agent(context);
 
    // Set planning strategy based on user choice
    AutonomousAgent::PlanningStrategy strategy;
    switch (strategy_choice) {
        case 2:
            strategy = AutonomousAgent::PlanningStrategy::PLAN_AND_EXECUTE;
            break;
        case 1:
        default:
            strategy = AutonomousAgent::PlanningStrategy::REACT;
    }
    agent.setPlanningStrategy(strategy);
 
    // Set the agent prompt (this extends the context system prompt for the agent)
    agent.setAgentPrompt(
        "You are an advanced autonomous assistant capable of using tools to help users "
        "accomplish their tasks. You break down complex problems into manageable steps "
        "and execute them systematically. Always provide clear explanations of your "
        "reasoning and approach."
    );
 
    // Set up options
    AutonomousAgent::Options agent_options;
    agent_options.max_iterations = 15;
 
    // Ask user if they want human-in-the-loop mode
    Logger::info("Enable human-in-the-loop mode? (y/n): ");
    char human_loop_choice;
    std::cin >> human_loop_choice;
    std::cin.ignore(); // Clear the newline
 
    agent_options.human_feedback_enabled = (human_loop_choice == 'y' || human_loop_choice == 'Y');
    if (agent_options.human_feedback_enabled) {
        agent_options.human_in_the_loop = detailedHumanApproval;
    }
 
    agent.setOptions(agent_options);
 
    // Set up callbacks
    agent.setStepCallback(detailedStepCallback);
 
    // Initialize the agent
    agent.init();
 
    // Get user input
    Logger::info("\n==================================================");
    Logger::info("                AUTONOMOUS AGENT                  ");
    Logger::info("==================================================");
    Logger::info("Enter a question or task for the agent (or 'exit' to quit):");
 
    String user_input;
    while (true) {
        Logger::info("\n> ");
        std::getline(std::cin, user_input);
 
        if (user_input == "exit" || user_input == "quit" || user_input == "q") {
            break;
        }
 
        if (user_input.empty()) {
            continue;
        }
 
        try {
            // Start a timer to measure execution time
            auto start_time = std::chrono::high_resolution_clock::now();
 
            // Run the agent
            JsonObject result = blockingWait(agent.run(user_input));
 
            // End timer
            auto end_time = std::chrono::high_resolution_clock::now();
            auto duration = std::chrono::duration_cast<std::chrono::seconds>(end_time - start_time).count();
 
            // Display the final result
            Logger::info("\n==================================================");
            Logger::info("                  FINAL RESULT                    ");
            Logger::info("==================================================");
            Logger::info("{}", result["answer"].get<String>());
 
            // Display completion statistics
            Logger::info("\n--------------------------------------------------");
            Logger::info("Task completed in {} seconds", duration);
            Logger::info("Total steps: {}", result["steps"].get<JsonArray>().size());
 
            if (result.contains("tool_calls")) {
                Logger::info("Tool calls: {}", result["tool_calls"].get<int>());
            }
 
            Logger::info("==================================================");
        } catch (const std::exception& e) {
            Logger::error("Error: {}", e.what());
        }
    }
 
    return EXIT_SUCCESS;
}