@decido/kernel-bridge 1.0.0 → 4.0.2

package/src/ai/services/LocalAgentResponder.ts

@@ -1,199 +0,0 @@
-/**
- * LocalAgentResponder — Local-first AI responses for DecidoOS
- * 
- * Handles user input locally when Cortex is not connected.
- * Provides greeting responses, tool execution, system info queries,
- * and basic conversational ability without requiring a remote AI.
- */
-
-
-// ─── Types ──────────────────────────────────────────────────
-
-interface LocalResponse {
-    text: string;
-    toolCalls?: Array<{ name: string; args: Record<string, unknown> }>;
-}
-
-// ─── Greeting Patterns ──────────────────────────────────────
-
-const GREETING_PATTERNS = [
-    /^(hola|hey|hi|hello|buenas|qué tal|que tal|saludos)\s*(decido|decidoos|agente|asistente)?/i,
-    /^(buenos?\s*(días|tardes|noches))/i,
-    /^(órale|oye|ey)\s*(decido)?/i,
-];
-
-const HELP_PATTERNS = [
-    /^(ayuda|help|qué puedes hacer|que puedes hacer|comandos|tools|herramientas)/i,
-    /^(qué|que)\s*(sabes|haces|puedes)/i,
-];
-
-const SYSTEM_PATTERNS = [
-    /^(estado|status|sistema|system)\s*(del\s*sistema)?/i,
-    /^(cómo|como)\s*(estás|estas|está|va)/i,
-];
-
-// ─── Response Templates ─────────────────────────────────────
-
-function getTimeGreeting(): string {
-    const hour = new Date().getHours();
-    if (hour < 12) return 'Buenos días';
-    if (hour < 18) return 'Buenas tardes';
-    return 'Buenas noches';
-}
-
-function getWelcomeMessage(): string {
-    const greeting = getTimeGreeting();
-
-    return `${greeting}, operador. Soy **DecidoOS Agent** 🧠
-
-Estoy en línea y operativo. Aquí está mi reporte de estado:
-
-▸ **Modo**: Local-first (sin dependencia de Cortex)
-▸ **Estado**: ✅ ONLINE
-
-Puedo auditar tu seguridad, escanear tu flota de procesos y más.
-
-Escribe **"ayuda"** para ver mis comandos principales.`;
-}
-
-function getHelpMessage(): string {
-    return `🛠️ **Capacidades de DecidoOS Agent**
-
-• "escanea puertos" — Auditar puertos abiertos
-• "escanea red" — Monitorear conexiones de red
-• "escanea flota" — Ver procesos Agent / Node / Python
-• "lista tareas" — Ver tareas / playbooks
-• "estado" — Estado del sistema
-
-**Voice Mode:**
-Presiona el botón verde 📞 para activar conversación por voz con Cortex.`;
-}
-
-async function getSystemStatus(): Promise<string> {
-    let cpuInfo = 'N/A';
-    let memInfo = 'N/A';
-
-    try {
-        const { invoke } = await import('@tauri-apps/api/core');
-        const cpu = await invoke<{ usage: number }>('get_cpu_usage');
-        const mem = await invoke<{ used: number; total: number }>('get_memory_usage');
-        cpuInfo = `${cpu.usage?.toFixed(1)}%`;
-        memInfo = `${(mem.used / 1024 / 1024 / 1024).toFixed(1)} GB / ${(mem.total / 1024 / 1024 / 1024).toFixed(1)} GB`;
-    } catch {
-        cpuInfo = '(requiere Tauri)';
-        memInfo = '(requiere Tauri)';
-    }
-
-    return `📊 **Estado del Sistema DecidoOS**
-
-▸ **CPU**: ${cpuInfo}
-▸ **RAM**: ${memInfo}
-▸ **Cortex**: Local mode
-▸ **Uptime Session**: ${getSessionUptime()}
-
-Todo operativo. ¿Necesitas algo más?`;
-}
-
-function getSessionUptime(): string {
-    const uptime = performance.now();
-    const minutes = Math.floor(uptime / 60000);
-    const hours = Math.floor(minutes / 60);
-    if (hours > 0) return `${hours}h ${minutes % 60}m`;
-    return `${minutes}m`;
-}
-
-// ─── Main Responder ─────────────────────────────────────────
-
-export async function processLocalMessage(input: string): Promise<LocalResponse> {
-    const trimmed = input.trim();
-
-    // 1. Greeting
-    if (GREETING_PATTERNS.some(p => p.test(trimmed))) {
-        return { text: getWelcomeMessage() };
-    }
-
-    // 2. Help
-    if (HELP_PATTERNS.some(p => p.test(trimmed))) {
-        return { text: getHelpMessage() };
-    }
-
-    // 3. System status
-    if (SYSTEM_PATTERNS.some(p => p.test(trimmed))) {
-        const status = await getSystemStatus();
-        return { text: status };
-    }
-
-    // 4. Tool execution patterns
-    const execMatch = trimmed.match(/^(ejecuta|run|exec|corre)\s+(.+)/i);
-    if (execMatch) {
-        const command = execMatch[2];
-        return {
-            text: `⚡ Ejecutando: \`${command}\`...`,
-            toolCalls: [{ name: 'tactical.execute_command', args: { command } }],
-        };
-    }
-
-    const scanMatch = trimmed.match(/^(escanea|scan|audita|audit)\s*(puertos|ports)/i);
-    if (scanMatch) {
-        return {
-            text: '🔍 Escaneando puertos del sistema...',
-            toolCalls: [{ name: 'security.audit_ports', args: { includeLoopback: false } }],
-        };
-    }
-
-    const networkMatch = trimmed.match(/^(escanea|scan|monitorea|monitor)\s*(red|network|conexiones|connections)/i);
-    if (networkMatch) {
-        return {
-            text: '🌐 Analizando conexiones de red...',
-            toolCalls: [{ name: 'security.network_monitor', args: {} }],
-        };
-    }
-
-    const fleetMatch = trimmed.match(/^(escanea|scan)\s*(flota|fleet|procesos|processes)/i);
-    if (fleetMatch) {
-        return {
-            text: '🛸 Escaneando flotilla de agentes...',
-            toolCalls: [{ name: 'tactical.scan_fleet', args: {} }],
-        };
-    }
-
-    const listMatch = trimmed.match(/^(lista|list|muestra|show)\s*(tareas|tasks|playbooks)/i);
-    if (listMatch) {
-        return {
-            text: '📋 Listando tareas...',
-            toolCalls: [{ name: 'tactical.list_tasks', args: { status: 'all' } }],
-        };
-    }
-
-    // 5. Forensic scan
-    const forensicMatch = trimmed.match(/^(analiza|analyze|forense|forensic)\s*(proceso|process)?\s*(\d+)?/i);
-    if (forensicMatch) {
-        const pid = forensicMatch[3] ? parseInt(forensicMatch[3]) : undefined;
-        return {
-            text: pid ? `🔬 Analizando proceso PID ${pid}...` : '🔬 Escaneando procesos sospechosos...',
-            toolCalls: [{ name: 'security.forensic_scan', args: { pid, deep: true } }],
-        };
-    }
-
-    // 6. Vulnerability scan
-    const vulnMatch = trimmed.match(/^(vulnerabilidades|vulnerabilities|vuln|audit\s*npm)/i);
-    if (vulnMatch) {
-        return {
-            text: '🛡️ Escaneando vulnerabilidades en dependencias...',
-            toolCalls: [{ name: 'security.scan_vulnerabilities', args: {} }],
-        };
-    }
-
-    // 7. Default - unrecognized
-    return {
-        text: `🤖 Entendido: "${trimmed.slice(0, 100)}"
-
-No tengo una respuesta local para esto. Para respuestas inteligentes con IA, conecta a **Cortex** (mindframe-cortex).
-
-Mientras tanto, prueba:
-• "ayuda" — ver mis capacidades
-• "ejecuta ls" — ejecutar un comando
-• "escanea puertos" — auditar seguridad
-• "estado" — ver estado del sistema`,
-    };
-}

package/src/ai/services/MLXBridge.ts

@@ -1,278 +0,0 @@
-/**
- * MLXBridge — Bridge between Tauri desktop app and MLX Python scripts
- * 
- * Spawns Python processes for:
- *  - Model inference (generate text, transcribe audio, generate images)
- *  - LoRA fine-tuning with live progress streaming
- *  - Model benchmarking and comparison
- * 
- * Uses Tauri shell plugin to manage Python child processes.
- */
-
-// Tauri core is loaded dynamically for browser compatibility
-
-// ─── Types ───────────────────────────────────────────────────
-
-export interface MLXModelInfo {
-    name: string;
-    family: string;       // 'llm' | 'vision' | 'audio' | 'image-gen'
-    paramCount?: string;   // e.g. "7B", "13B"
-    quantization?: string; // e.g. "4bit", "8bit"
-    path: string;          // HuggingFace path or local
-    loaded: boolean;
-}
-
-export interface InferenceResult {
-    text: string;
-    tokensPerSecond: number;
-    totalTokens: number;
-    latencyMs: number;
-    model: string;
-}
-
-export interface TrainingProgress {
-    epoch: number;
-    totalEpochs: number;
-    step: number;
-    totalSteps: number;
-    loss: number;
-    learningRate: number;
-    tokensPerSecond: number;
-    elapsedMs: number;
-}
-
-export interface BenchmarkResult {
-    model: string;
-    promptTokens: number;
-    generatedTokens: number;
-    tokensPerSecond: number;
-    latencyMs: number;
-    memoryMb: number;
-}
-
-// ─── Helper: Run Shell Command ───────────────────────────────
-
-async function runShellCommand(cmd: string, args: string[]): Promise<string> {
-    try {
-        const { invoke } = await import('@tauri-apps/api/core');
-        const result = await invoke<string>('run_shell_command', {
-            command: cmd,
-            args,
-        });
-        return result;
-    } catch (err) {
-        console.error('[MLXBridge] Shell command failed:', err);
-        throw err;
-    }
-}
-
-// ─── Ollama API (extends OllamaService) ──────────────────────
-
-const OLLAMA_URL = 'http://localhost:11434';
-
-async function ollamaChat(model: string, prompt: string, temperature = 0.7): Promise<InferenceResult> {
-    const start = Date.now();
-    try {
-        const res = await fetch(`${OLLAMA_URL}/api/chat`, {
-            method: 'POST',
-            headers: { 'Content-Type': 'application/json' },
-            body: JSON.stringify({
-                model,
-                messages: [{ role: 'user', content: prompt }],
-                stream: false,
-                options: { temperature },
-            }),
-        });
-        const data = await res.json();
-        const latencyMs = Date.now() - start;
-        return {
-            text: data.message?.content ?? '',
-            tokensPerSecond: data.eval_count ? (data.eval_count / (latencyMs / 1000)) : 0,
-            totalTokens: data.eval_count ?? 0,
-            latencyMs,
-            model,
-        };
-    } catch {
-        return { text: 'Error: Ollama no disponible', tokensPerSecond: 0, totalTokens: 0, latencyMs: Date.now() - start, model };
-    }
-}
-
-async function ollamaListModels(): Promise<string[]> {
-    try {
-        const res = await fetch(`${OLLAMA_URL}/api/tags`);
-        if (!res.ok) return [];
-        const data = await res.json();
-        return (data.models || []).map((m: { name: string }) => m.name);
-    } catch {
-        return [];
-    }
-}
-
-async function ollamaPull(model: string): Promise<void> {
-    await fetch(`${OLLAMA_URL}/api/pull`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({ name: model }),
-    });
-}
-
-// ─── MLX Python Bridge ──────────────────────────────────────
-
-async function mlxGenerate(model: string, prompt: string): Promise<InferenceResult> {
-    const start = Date.now();
-    try {
-        const output = await runShellCommand('python3', [
-            '-m', 'mlx_lm.generate',
-            '--model', model,
-            '--prompt', prompt,
-            '--max-tokens', '512',
-        ]);
-        return {
-            text: output,
-            tokensPerSecond: 0, // parsed from output in production
-            totalTokens: output.split(' ').length,
-            latencyMs: Date.now() - start,
-            model,
-        };
-    } catch (err) {
-        return { text: `MLX Error: ${err}`, tokensPerSecond: 0, totalTokens: 0, latencyMs: Date.now() - start, model };
-    }
-}
-
-// ─── Public API ──────────────────────────────────────────────
-
-/** List all available models (Ollama + MLX catalog) */
-export async function listAvailableModels(): Promise<MLXModelInfo[]> {
-    const models: MLXModelInfo[] = [];
-
-    // Ollama models
-    const ollamaModels = await ollamaListModels();
-    for (const name of ollamaModels) {
-        models.push({
-            name,
-            family: 'llm',
-            path: `ollama:${name}`,
-            loaded: true,
-        });
-    }
-
-    // MLX model catalog (known models from decido-mlx)
-    const mlxCatalog: MLXModelInfo[] = [
-        { name: 'Qwen2.5-7B-Instruct-4bit', family: 'llm', paramCount: '7B', quantization: '4bit', path: 'mlx-community/Qwen2.5-7B-Instruct-4bit', loaded: false },
-        { name: 'Mistral-7B-Instruct-v0.3-4bit', family: 'llm', paramCount: '7B', quantization: '4bit', path: 'mlx-community/Mistral-7B-Instruct-v0.3-4bit', loaded: false },
-        { name: 'Llama-3.2-3B-Instruct-4bit', family: 'llm', paramCount: '3B', quantization: '4bit', path: 'mlx-community/Llama-3.2-3B-Instruct-4bit', loaded: false },
-        { name: 'Mixtral-8x7B-Instruct-v0.1-4bit', family: 'llm', paramCount: '46.7B', quantization: '4bit', path: 'mlx-community/Mixtral-8x7B-Instruct-v0.1-4bit', loaded: false },
-        { name: 'Whisper-large-v3', family: 'audio', path: 'mlx-community/whisper-large-v3', loaded: false },
-        { name: 'CLIP-ViT-B-32', family: 'vision', path: 'openai/clip-vit-base-patch32', loaded: false },
-        { name: 'FLUX.1-schnell-4bit', family: 'image-gen', path: 'mlx-community/FLUX.1-schnell-4bit-quantized', loaded: false },
-        { name: 'Stable-Diffusion-XL', family: 'image-gen', path: 'mlx-community/sdxl-turbo', loaded: false },
-    ];
-    models.push(...mlxCatalog);
-
-    return models;
-}
-
-/** Run inference on a model (auto-routes Ollama vs MLX) */
-export async function runInference(
-    modelPath: string,
-    prompt: string,
-    options?: { temperature?: number }
-): Promise<InferenceResult> {
-    if (modelPath.startsWith('ollama:')) {
-        const model = modelPath.replace('ollama:', '');
-        return ollamaChat(model, prompt, options?.temperature);
-    }
-    return mlxGenerate(modelPath, prompt);
-}
-
-/** Compare two models side-by-side */
-export async function compareModels(
-    modelA: string,
-    modelB: string,
-    prompt: string
-): Promise<{ a: InferenceResult; b: InferenceResult }> {
-    const [a, b] = await Promise.all([
-        runInference(modelA, prompt),
-        runInference(modelB, prompt),
-    ]);
-    return { a, b };
-}
-
-/** Pull/download a model */
-export async function pullModel(model: string): Promise<void> {
-    if (model.startsWith('ollama:')) {
-        await ollamaPull(model.replace('ollama:', ''));
-    } else {
-        // MLX models via huggingface-cli
-        await runShellCommand('huggingface-cli', ['download', model]);
-    }
-}
-
-/** Benchmark a model with standard prompts */
-export async function benchmarkModel(modelPath: string): Promise<BenchmarkResult> {
-    const testPrompts = [
-        'Explain quantum computing in one paragraph.',
-        'Write a Python function to sort a list.',
-        'What is the capital of Colombia?',
-    ];
-    let totalTokens = 0;
-    let totalMs = 0;
-
-    for (const prompt of testPrompts) {
-        const result = await runInference(modelPath, prompt);
-        totalTokens += result.totalTokens;
-        totalMs += result.latencyMs;
-    }
-
-    return {
-        model: modelPath,
-        promptTokens: testPrompts.join(' ').split(' ').length,
-        generatedTokens: totalTokens,
-        tokensPerSecond: totalTokens / (totalMs / 1000),
-        latencyMs: totalMs / testPrompts.length,
-        memoryMb: 0, // would need system metrics
-    };
-}
-
-/** Start LoRA training — returns a handle to monitor progress */
-export function startLoRATraining(config: {
-    baseModel: string;
-    dataPath: string;
-    outputPath: string;
-    epochs?: number;
-    batchSize?: number;
-    learningRate?: number;
-    loraLayers?: number;
-}): { stop: () => void; onProgress: (cb: (p: TrainingProgress) => void) => void } {
-    let stopped = false;
-    let progressCallback: ((p: TrainingProgress) => void) | null = null;
-
-    // Simulate progress (in production, parse Python stdout)
-    const totalSteps = (config.epochs ?? 10) * 100;
-    let step = 0;
-    const start = Date.now();
-
-    const interval = setInterval(() => {
-        if (stopped || step >= totalSteps) {
-            clearInterval(interval);
-            return;
-        }
-        step++;
-        const progress: TrainingProgress = {
-            epoch: Math.floor(step / 100) + 1,
-            totalEpochs: config.epochs ?? 10,
-            step,
-            totalSteps,
-            loss: 2.5 * Math.exp(-step / 200) + 0.3 + Math.random() * 0.1,
-            learningRate: config.learningRate ?? 1e-5,
-            tokensPerSecond: 150 + Math.random() * 50,
-            elapsedMs: Date.now() - start,
-        };
-        progressCallback?.(progress);
-    }, 500);
-
-    return {
-        stop: () => { stopped = true; clearInterval(interval); },
-        onProgress: (cb) => { progressCallback = cb; },
-    };
-}