Wi-Fi Troubleshooter Agent

Source Code: cpp/examples/wifi_agent.cpp — single-file, self-contained agent (~1,100 lines including 13 tools and a custom TUI).

Platform: Windows (PowerShell network commands). Compiles on Linux for CI but tools require Windows to return real data. Prerequisite: Lemonade Server running with a model loaded.

What This Agent Does

The Wi-Fi Troubleshooter is an AI agent that acts like an IT support specialist. When a user says “my internet isn’t working”, it runs a systematic diagnostic chain, reads real output from your machine, reasons about what’s wrong, and applies fixes automatically.

Runs real commands — every tool executes an actual PowerShell command and returns real output
LLM-driven reasoning — no hardcoded if/else tree; the LLM reads each result and decides the next step
Pure C++ — no Python, no MCP subprocess, no external dependencies. Just a compiled binary talking to a local LLM
Fully local — the LLM runs on your AMD hardware via Lemonade Server. No data leaves your machine

Demo

Full diagnostic run on a Ryzen AI PC: the agent calls 5 tools in sequence, reads real network output, and reports status — all powered by a local LLM on the NPU.

Quick Start

Build

Windows (MSVC)
Windows (Ninja)
Linux

cd cpp

cmake -B build -G "Visual Studio 17 2022" -A x64

cmake --build build --config Release

Binary: cpp\build\Release\wifi_agent.exe

cd cpp

cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=Release

cmake --build build

cd cpp

cmake -B build -DCMAKE_BUILD_TYPE=Release

cmake --build build

Compiles on Linux for CI purposes, but the PowerShell-based tools require Windows to return meaningful results.

Start Lemonade Server

lemonade-server serve

The agent connects to http://localhost:8000/api/v1 by default.

Run the agent (run as admin for fix tools)

cpp\build\Release\wifi_agent.exe

Try these prompts:

You: Run a full network diagnostic.
You: My Wi-Fi is connected but I can't browse the web.
You: Check if DNS is working and fix it if not.

Architecture

The agent loop is a conversation between your C++ code and the LLM: Your C++ code provides the tools (what the agent can do), the system prompt provides the strategy (what it should do), and the LLM connects the two.

How It Works

The agent subclasses gaia::Agent with three pieces: a config, registered tools, and a system prompt.

class WiFiTroubleshooterAgent : public gaia::Agent {
public:
    explicit WiFiTroubleshooterAgent(const std::string& modelId)
        : Agent(makeConfig(modelId)) {
        setOutputHandler(std::make_unique<CleanConsole>());  // custom TUI
        init();  // calls registerTools() + composes system prompt
    }

protected:
    std::string getSystemPrompt() const override {
        return R"(You are an expert Windows network troubleshooter...
            // Diagnostic sequence, fix instructions, output format
        )";
    }

    void registerTools() override {
        // 13 tools: 7 diagnostic + 6 fix (see tables below)
        toolRegistry().registerTool("check_adapter", ...);
        toolRegistry().registerTool("ping_host", ...);
        // ...
    }

private:
    static gaia::AgentConfig makeConfig(const std::string& modelId) {
        gaia::AgentConfig config;
        config.maxSteps = 20;   // up to 20 tool calls per query
        config.modelId = modelId;
        return config;
    }
};

Each tool is a C++ lambda that runs a PowerShell command and returns the output:

toolRegistry().registerTool(
    "check_adapter",                          // name the LLM calls
    "Show Wi-Fi adapter status including "    // description the LLM reads
    "SSID, signal strength, and state.",
    [](const gaia::json& /*args*/) -> gaia::json {
        std::string output = runShell("netsh wlan show interfaces");
        return {{"tool", "check_adapter"}, {"output", output}};
    },
    {}  // no parameters
);

Tools with parameters validate input before passing to the shell:

toolRegistry().registerTool(
    "ping_host",
    "Ping a specific host and return connection status.",
    [](const gaia::json& args) -> gaia::json {
        std::string host = args.value("host", "");
        if (host.empty()) return {{"error", "host parameter is required"}};
        if (!isSafeShellArg(host)) return {{"error", "Invalid host"}};
        std::string output = runShell("Test-NetConnection -ComputerName " + host + " | ConvertTo-Json");
        return {{"tool", "ping_host"}, {"output", output}};
    },
    {{"host", gaia::ToolParamType::STRING, true, "Hostname or IP to ping"}}
);

All string arguments from the LLM are validated with isSafeShellArg() to reject shell metacharacters before constructing commands. See the Custom Agent guide for details on tool registration patterns.

Tool Reference

Diagnostic Tools (Read-Only)

Tool	What It Runs	Parameters
`check_adapter`	`netsh wlan show interfaces`	none
`check_wifi_drivers`	`netsh wlan show drivers`	none
`check_ip_config`	`ipconfig /all`	none
`test_dns_resolution`	`Resolve-DnsName`	`hostname` (optional)
`test_internet`	`Test-NetConnection 8.8.8.8 -Port 443`	none
`test_bandwidth`	Cloudflare CDN parallel download/upload	none
`ping_host`	`Test-NetConnection <host>`	`host` (required)

Fix Tools (Write Operations)

Tool	What It Runs	When to Use
`flush_dns_cache`	`Clear-DnsClientCache`	Stale DNS entries
`set_dns_servers`	`Set-DnsClientServerAddress`	Default DNS not responding
`renew_dhcp_lease`	`ipconfig /release` + `/renew`	No IP or APIPA address
`toggle_wifi_radio`	Windows Radio Management API	Adapter shows “Software Off”
`enable_wifi_adapter`	`Enable-NetAdapter`	Adapter administratively disabled
`restart_wifi_adapter`	`Disable-NetAdapter` + `Enable-NetAdapter`	Adapter stuck in bad state

Diagnostic Flow

The system prompt teaches the LLM this decision tree. Each fix loops back to re-verify before declaring success:

Extending the Agent

Add a tool inside registerTools() and update the system prompt:

toolRegistry().registerTool(
    "check_firewall",
    "Check if Windows Firewall is blocking outbound connections.",
    [](const gaia::json& /*args*/) -> gaia::json {
        std::string output = runShell(
            "Get-NetFirewallProfile | Select-Object Name, Enabled | ConvertTo-Json"
        );
        return {{"tool", "check_firewall"}, {"output", output}};
    },
    {}
);

The framework handles everything else — the new tool appears in the LLM’s system prompt automatically.

Next Steps

System Health Agent

Compare with the MCP-based approach for system diagnostics

C++ Framework Overview

AgentConfig reference, project structure, and how the agent loop works

Integration Guide

Use gaia_core in your own CMake project via FetchContent or find_package

Customizing Your Agent

Custom prompts, typed tools, MCP servers, output capture, and tuning

Getting Started

C++ Framework

Python Framework

Wi-Fi Troubleshooter Agent

What This Agent Does

Demo

Quick Start

Architecture

How It Works

Tool Reference

Diagnostic Tools (Read-Only)

Fix Tools (Write Operations)

Diagnostic Flow

Extending the Agent

Next Steps

System Health Agent

C++ Framework Overview

Integration Guide

Customizing Your Agent

Getting Started

C++ Framework

Python Framework

​What This Agent Does

​Demo

​Quick Start

​Architecture

​How It Works

​Tool Reference

​Diagnostic Tools (Read-Only)

​Fix Tools (Write Operations)

​Diagnostic Flow

​Extending the Agent

​Next Steps

System Health Agent

C++ Framework Overview

Integration Guide

Customizing Your Agent

What This Agent Does

Demo

Quick Start

Architecture

How It Works

Tool Reference

Diagnostic Tools (Read-Only)

Fix Tools (Write Operations)

Diagnostic Flow

Extending the Agent

Next Steps