@appkit/llamacpp-cli 1.11.0 → 1.12.1

package/src/tui/SplashScreen.ts

@@ -1,149 +0,0 @@
-import blessed from "blessed";
-import { readFileSync } from "fs";
-import { join } from "path";
-
-// Get version from package.json at runtime
-function getVersion(): string {
-  try {
-    // Try to find package.json relative to this file's location
-    // Works both in src/ (dev) and dist/ (production)
-    const possiblePaths = [
-      join(__dirname, "../../package.json"), // From dist/tui/
-      join(__dirname, "../../../package.json"), // Fallback
-    ];
-
-    for (const pkgPath of possiblePaths) {
-      try {
-        const content = readFileSync(pkgPath, "utf-8");
-        const pkg = JSON.parse(content);
-        if (pkg.version) return pkg.version;
-      } catch {
-        continue;
-      }
-    }
-    return "1.0.0";
-  } catch {
-    return "1.0.0";
-  }
-}
-
-// ASCII art logo for llama.cpp
-const LOGO = `
- {cyan-fg} ██╗     ██╗      █████╗ ███╗   ███╗ █████╗    ██████╗██████╗ ██████╗{/cyan-fg}
- {cyan-fg}██║     ██║     ██╔══██╗████╗ ████║██╔══██╗  ██╔════╝██╔══██╗██╔══██╗{/cyan-fg}
- {cyan-fg}██║     ██║     ███████║██╔████╔██║███████║  ██║     ██████╔╝██████╔╝{/cyan-fg}
- {cyan-fg}██║     ██║     ██╔══██║██║╚██╔╝██║██╔══██║  ██║     ██╔═══╝ ██╔═══╝{/cyan-fg}
- {cyan-fg}███████╗███████╗██║  ██║██║ ╚═╝ ██║██║  ██║  ╚██████╗██║     ██║{/cyan-fg}
- {cyan-fg}╚══════╝╚══════╝╚═╝  ╚═╝╚═╝     ╚═╝╚═╝  ╚═╝   ╚═════╝╚═╝     ╚═╝{/cyan-fg}
-`.trim();
-
-export interface SplashCallbacks {
-  onLoadConfigs: () => Promise<void>;
-  onCheckServices: () => Promise<void>;
-  onInitMetrics: () => Promise<void>;
-}
-
-export interface SplashScreenControls {
-  updateStatus: (line: number, text: string, done?: boolean) => void;
-  complete: () => void;
-}
-
-/**
- * Creates and displays the splash screen with loading status
- * Returns a cleanup function to remove the splash when the main UI is ready
- */
-export async function createSplashScreen(
-  screen: blessed.Widgets.Screen,
-  callbacks: SplashCallbacks,
-): Promise<() => void> {
-  const version = getVersion();
-
-  // Create top-left aligned container
-  const container = blessed.box({
-    top: 0,
-    left: 0,
-    width: 75,
-    height: 19,
-    tags: true,
-  });
-  screen.append(container);
-
-  // Logo box
-  const logoBox = blessed.box({
-    parent: container,
-    top: 0,
-    left: 0,
-    width: 73,
-    height: 7,
-    tags: true,
-    content: LOGO,
-  });
-
-  // Info box with border
-  const infoBox = blessed.box({
-    parent: container,
-    top: 7,
-    left: 0,
-    width: 73,
-    height: 10,
-    tags: true,
-    border: {
-      type: "line",
-    },
-    style: {
-      border: {
-        fg: "blue",
-      },
-    },
-  });
-
-  // Status lines (inside the info box)
-  const statusLines = [
-    "{bold}Local LLM Server Manager{/bold}                                    v" +
-      version,
-    "",
-    "{gray-fg}>{/gray-fg} Loading server configurations...",
-    "{gray-fg}>{/gray-fg} Checking launchctl services...",
-    "{gray-fg}>{/gray-fg} Initializing metrics collectors...",
-    "{gray-fg}>{/gray-fg} Loading UI...",
-  ];
-
-  function updateDisplay() {
-    infoBox.setContent(statusLines.join("\n"));
-    screen.render();
-  }
-
-  function updateStatus(line: number, text: string, done: boolean = false) {
-    const prefix = done ? "{green-fg}✓{/green-fg}" : "{yellow-fg}>{/yellow-fg}";
-    statusLines[line + 2] = `${prefix} ${text}`;
-    updateDisplay();
-  }
-
-  // Initial render
-  updateDisplay();
-
-  // Run loading sequence
-  // Step 1: Load configs
-  updateStatus(0, "Loading server configurations...", false);
-  await callbacks.onLoadConfigs();
-  updateStatus(0, "Server configurations loaded", true);
-
-  // Step 2: Check services
-  updateStatus(1, "Checking launchctl services...", false);
-  await callbacks.onCheckServices();
-  updateStatus(1, "Launchctl services checked", true);
-
-  // Step 3: Init metrics
-  updateStatus(2, "Initializing metrics collectors...", false);
-  await callbacks.onInitMetrics();
-  updateStatus(2, "Metrics collectors ready", true);
-
-  // Step 4: Loading UI (stays active until cleanup is called)
-  updateStatus(3, "Loading UI...", false);
-
-  // Return cleanup function - caller decides when to remove splash
-  return () => {
-    screen.remove(container);
-    screen.render();
-  };
-}

package/src/types/global-config.ts

@@ -1,26 +0,0 @@
-export interface GlobalConfig {
-  version: string;
-  defaultPort: number;
-  modelsDirectory: string;      // ~/models expanded to full path
-  llamaServerBinary: string;    // /opt/homebrew/bin/llama-server
-  defaults: {
-    threads: number;
-    ctxSize: number;
-    gpuLayers: number;
-  };
-}
-
-/**
- * Default global configuration
- */
-export const DEFAULT_GLOBAL_CONFIG: GlobalConfig = {
-  version: '1.0.0',
-  defaultPort: 9000,
-  modelsDirectory: '',  // Set at runtime
-  llamaServerBinary: '/opt/homebrew/bin/llama-server',
-  defaults: {
-    threads: 8,
-    ctxSize: 8192,
-    gpuLayers: 60,
-  },
-};

package/src/types/history-types.ts

@@ -1,39 +0,0 @@
-// Historical monitoring data types
-
-export interface HistorySnapshot {
-  timestamp: number; // Unix timestamp in milliseconds
-  server: {
-    healthy: boolean;
-    uptime?: string;
-    activeSlots: number;
-    idleSlots: number;
-    totalSlots: number;
-    avgPromptSpeed?: number;       // Tokens per second
-    avgGenerateSpeed?: number;     // Tokens per second
-    processMemory?: number;        // Bytes (RSS)
-    processCpuUsage?: number;      // Percentage (0-100+) from ps
-  };
-  system?: {
-    gpuUsage?: number;             // Percentage (0-100)
-    cpuUsage?: number;             // Percentage (0-100)
-    aneUsage?: number;             // Percentage (0-100)
-    temperature?: number;          // Celsius
-    memoryUsed: number;            // Bytes
-    memoryTotal: number;           // Bytes
-  };
-}
-
-export interface HistoryData {
-  serverId: string;
-  snapshots: HistorySnapshot[];
-}
-
-export type TimeWindow = '1h' | '6h' | '24h';
-
-export const TIME_WINDOW_HOURS: Record<TimeWindow, number> = {
-  '1h': 1,
-  '6h': 6,
-  '24h': 24,
-};
-
-export const TIME_WINDOWS: TimeWindow[] = ['1h', '6h', '24h'];

package/src/types/model-info.ts

@@ -1,8 +0,0 @@
-export interface ModelInfo {
-  filename: string;         // Original filename
-  path: string;             // Full absolute path
-  size: number;             // File size in bytes
-  sizeFormatted: string;    // Human-readable size (e.g., "1.9 GB")
-  modified: Date;           // Last modified date
-  exists: boolean;          // File exists and is readable
-}

package/src/types/monitor-types.ts

@@ -1,162 +0,0 @@
-import { ServerConfig } from './server-config.js';
-
-// llama.cpp API response types
-
-export interface HealthResponse {
-  status: string;
-  error?: string;
-}
-
-export interface PropsResponse {
-  default_generation_settings: {
-    n_ctx: number;
-    n_predict: number;
-    model: string;
-    seed: number;
-    temperature: number;
-    top_k: number;
-    top_p: number;
-    min_p: number;
-    n_keep: number;
-    stream: boolean;
-  };
-  total_slots: number;
-  model_loaded: boolean;
-  model_path: string;
-  model_alias?: string;
-}
-
-export interface SlotInfo {
-  id: number;
-  state: 'idle' | 'processing';
-  task_id?: number;
-  prompt?: string;
-  n_prompt_tokens?: number;
-  n_decoded?: number;
-  n_ctx: number;
-  truncated?: boolean;
-  stopped_eos?: boolean;
-  stopped_word?: boolean;
-  stopped_limit?: boolean;
-  stopping_word?: string;
-  tokens_predicted?: number;
-  tokens_evaluated?: number;
-  generation_settings?: {
-    n_ctx: number;
-    n_predict: number;
-    seed: number;
-    temperature: number;
-    top_k: number;
-    top_p: number;
-  };
-  prompt_tokens_processed?: number;
-  t_prompt_processing?: number;  // Time in ms
-  t_token_generation?: number;     // Time in ms
-  timings?: {
-    prompt_n: number;
-    prompt_ms: number;
-    prompt_per_token_ms: number;
-    prompt_per_second: number;
-    predicted_n: number;
-    predicted_ms: number;
-    predicted_per_token_ms: number;
-    predicted_per_second: number;
-  };
-}
-
-export interface SlotsResponse {
-  slots: SlotInfo[];
-}
-
-// System metrics types
-
-export interface SystemMetrics {
-  // GPU/CPU/ANE (from macmon if available)
-  gpuUsage?: number;          // Percentage (0-100)
-  cpuUsage?: number;          // Percentage (0-100)
-  cpuCores?: number;          // Number of cores
-  aneUsage?: number;          // Apple Neural Engine percentage (0-100)
-  temperature?: number;        // GPU temperature in Celsius
-
-  // Memory (from vm_stat or macmon)
-  memoryUsed: number;         // Bytes
-  memoryTotal: number;        // Bytes
-  swapUsed?: number;          // Bytes
-  processMemory?: number;     // Bytes (specific to llama-server process)
-
-  // Metadata
-  timestamp: number;
-  source: 'macmon' | 'vm_stat' | 'none';
-  warnings?: string[];        // e.g., "macmon not available, showing memory only"
-}
-
-// Aggregated metrics for TUI display
-
-export interface ServerMetrics {
-  // Server identification
-  server: ServerConfig;
-
-  // Health status
-  healthy: boolean;
-  uptime?: string;            // Human-readable (e.g., "2h 34m 12s")
-  error?: string;
-
-  // Model information
-  modelLoaded: boolean;
-  modelName: string;
-  contextSize: number;
-  totalSlots: number;
-
-  // Request metrics
-  activeSlots: number;
-  idleSlots: number;
-  slots: SlotInfo[];
-
-  // Performance metrics (derived from slots)
-  avgPromptSpeed?: number;    // Tokens per second
-  avgGenerateSpeed?: number;  // Tokens per second
-  requestsPerMinute?: number; // Estimated from slot activity
-  avgLatency?: number;        // Milliseconds
-
-  // Cache metrics (if available from /metrics endpoint)
-  cacheHitRate?: number;      // Percentage
-
-  // Process metrics
-  processMemory?: number;     // Bytes (actual RSS from top command)
-  processCpuUsage?: number;   // Percentage (0-100+) from ps command
-
-  // Timestamp
-  timestamp: number;
-  stale: boolean;             // True if data is from last successful fetch
-}
-
-export interface MonitorData {
-  server: ServerMetrics;
-  system?: SystemMetrics;
-  lastUpdated: Date;
-  updateInterval: number;     // Milliseconds
-  consecutiveFailures: number;
-}
-
-// Error and loading states
-
-export interface ErrorState {
-  error: string;
-  canRetry: boolean;
-  suggestions?: string[];
-}
-
-export interface LoadingState {
-  message: string;
-  progress?: number;          // 0-100 if determinate
-}
-
-// Collection result (for graceful degradation)
-
-export interface CollectionResult<T> {
-  success: boolean;
-  data: T | null;
-  error?: string;
-  warnings?: string[];
-  stale?: boolean;
-}

package/src/types/router-config.ts

@@ -1,25 +0,0 @@
-export type RouterStatus = 'running' | 'stopped' | 'crashed';
-
-export interface RouterConfig {
-  id: 'router';
-  port: number;
-  host: string;
-
-  // State tracking
-  status: RouterStatus;
-  pid?: number;
-  createdAt: string;
-  lastStarted?: string;
-  lastStopped?: string;
-
-  // launchctl metadata
-  plistPath: string;
-  label: 'com.llama.router';
-  stdoutPath: string;
-  stderrPath: string;
-
-  // Router settings
-  healthCheckInterval: number;  // ms between health checks (default: 5000)
-  requestTimeout: number;        // ms for backend requests (default: 120000)
-  verbose: boolean;              // Enable verbose logging to file (default: false)
-}

package/src/types/server-config.ts

@@ -1,46 +0,0 @@
-export type ServerStatus = 'running' | 'stopped' | 'crashed';
-
-export interface ServerConfig {
-  id: string;              // Sanitized model name (unique identifier)
-  modelPath: string;       // Full path to GGUF file
-  modelName: string;       // Display name (original filename)
-  port: number;            // Server port
-  host: string;            // Bind address (default: 127.0.0.1)
-
-  // llama-server configuration
-  threads: number;
-  ctxSize: number;
-  gpuLayers: number;
-  embeddings: boolean;     // Always true
-  jinja: boolean;          // Always true
-  verbose: boolean;        // Enable verbose HTTP logging (--log-verbose flag)
-  customFlags?: string[];  // Additional llama-server flags (e.g., ["--pooling", "mean"])
-
-  // State tracking
-  status: ServerStatus;
-  pid?: number;
-  createdAt: string;       // ISO timestamp
-  lastStarted?: string;    // ISO timestamp
-  lastStopped?: string;    // ISO timestamp
-  metalMemoryMB?: number;  // Metal (GPU) memory allocated in MB (parsed from logs)
-
-  // launchctl metadata
-  plistPath: string;       // Full path to plist file
-  label: string;           // launchctl service label (com.llama.<id>)
-
-  // Logging
-  stdoutPath: string;      // Path to stdout log
-  stderrPath: string;      // Path to stderr log
-}
-
-/**
- * Sanitize a model filename to create a valid server ID
- * Example: "llama-3.2-3b-instruct-q4_k_m.gguf" → "llama-3-2-3b-instruct-q4-k-m"
- */
-export function sanitizeModelName(modelName: string): string {
-  return modelName
-    .replace(/\.gguf$/i, '')           // Remove .gguf extension
-    .replace(/[^a-zA-Z0-9]+/g, '-')    // Replace non-alphanumeric with hyphens
-    .toLowerCase()                      // Lowercase
-    .replace(/^-+|-+$/g, '');          // Trim hyphens from ends
-}

package/src/utils/downsample-utils.ts

@@ -1,128 +0,0 @@
-/**
- * Downsampling utilities for time-series chart data
- * Uses time-aligned buckets to ensure stable charts as new data arrives
- */
-
-export interface TimeSeriesPoint {
-  timestamp: number;
-  value: number;
-}
-
-type AggregationMethod = 'max' | 'mean';
-
-const ONE_HOUR_MS = 60 * 60 * 1000;
-
-/**
- * Core bucketing logic shared by all downsampling functions.
- * Uses ABSOLUTE bucket boundaries that never shift, ensuring chart stability.
- */
-function createTimeBuckets(
-  data: TimeSeriesPoint[],
-  targetPoints: number,
-  startTime: number,
-  endTime: number
-): number[][] {
-  const timeRange = endTime - startTime;
-  const bucketDuration = Math.ceil(timeRange / targetPoints);
-  const alignedStart = Math.floor(startTime / bucketDuration) * bucketDuration;
-  const buckets: number[][] = Array.from({ length: targetPoints }, () => []);
-
-  for (const point of data) {
-    if (point.timestamp < startTime || point.timestamp > endTime) continue;
-    const bucketIndex = Math.floor((point.timestamp - alignedStart) / bucketDuration);
-    if (bucketIndex >= 0 && bucketIndex < targetPoints) {
-      buckets[bucketIndex].push(point.value);
-    }
-  }
-
-  return buckets;
-}
-
-/**
- * Aggregate bucket values using the specified method.
- */
-function aggregateBuckets(buckets: number[][], method: AggregationMethod): number[] {
-  return buckets.map(bucket => {
-    const validValues = method === 'max'
-      ? bucket.filter(v => !isNaN(v) && v > 0)
-      : bucket.filter(v => !isNaN(v) && isFinite(v));
-
-    if (validValues.length === 0) return 0;
-
-    if (method === 'max') {
-      return Math.max(...validValues);
-    }
-    return validValues.reduce((sum, v) => sum + v, 0) / validValues.length;
-  });
-}
-
-/**
- * Downsample using time-aligned bucket max - preserves peaks.
- * Best for: GPU/CPU usage, token speeds where peaks matter.
- */
-export function downsampleMaxTime(data: TimeSeriesPoint[], targetPoints: number): number[] {
-  if (data.length === 0) return [];
-  if (data.length <= targetPoints) return data.map(d => d.value);
-
-  const buckets = createTimeBuckets(
-    data,
-    targetPoints,
-    data[0].timestamp,
-    data[data.length - 1].timestamp
-  );
-  return aggregateBuckets(buckets, 'max');
-}
-
-/**
- * Downsample using time-aligned bucket mean - preserves average trends.
- * Best for: Memory usage where average is meaningful.
- */
-export function downsampleMeanTime(data: TimeSeriesPoint[], targetPoints: number): number[] {
-  if (data.length === 0) return [];
-  if (data.length <= targetPoints) return data.map(d => d.value);
-
-  const buckets = createTimeBuckets(
-    data,
-    targetPoints,
-    data[0].timestamp,
-    data[data.length - 1].timestamp
-  );
-  return aggregateBuckets(buckets, 'mean');
-}
-
-/**
- * Calculate downsampling ratio as a display string.
- */
-export function getDownsampleRatio(originalCount: number, targetCount: number): string {
-  if (originalCount <= targetCount) return '1:1';
-  const ratio = Math.round(originalCount / targetCount);
-  return `${ratio}:1`;
-}
-
-/**
- * Downsample with full hour coverage using max aggregation.
- * Creates buckets for the entire hour (60 minutes), filling gaps with 0.
- * Best for: Hour view where we want to show the full time range.
- */
-export function downsampleMaxTimeWithFullHour(data: TimeSeriesPoint[], targetPoints: number): number[] {
-  if (data.length === 0) return Array(targetPoints).fill(0);
-
-  const now = Date.now();
-  const oneHourAgo = now - ONE_HOUR_MS;
-  const buckets = createTimeBuckets(data, targetPoints, oneHourAgo, now);
-  return aggregateBuckets(buckets, 'max');
-}
-
-/**
- * Downsample with full hour coverage using mean aggregation.
- * Creates buckets for the entire hour (60 minutes), filling gaps with 0.
- * Best for: Hour view where we want to show the full time range.
- */
-export function downsampleMeanTimeWithFullHour(data: TimeSeriesPoint[], targetPoints: number): number[] {
-  if (data.length === 0) return Array(targetPoints).fill(0);
-
-  const now = Date.now();
-  const oneHourAgo = now - ONE_HOUR_MS;
-  const buckets = createTimeBuckets(data, targetPoints, oneHourAgo, now);
-  return aggregateBuckets(buckets, 'mean');
-}