@appkit/llamacpp-cli 1.4.1 → 1.6.0

package/dist/utils/process-utils.js

@@ -6,7 +6,11 @@ exports.execCommandFull = execCommandFull;
 exports.commandExists = commandExists;
 exports.isProcessRunning = isProcessRunning;
 exports.isPortInUse = isPortInUse;
+exports.spawnAndReadOneLine = spawnAndReadOneLine;
+exports.getBatchProcessMemory = getBatchProcessMemory;
 exports.getProcessMemory = getProcessMemory;
+exports.getBatchProcessCpu = getBatchProcessCpu;
+exports.getProcessCpu = getProcessCpu;
 const child_process_1 = require("child_process");
 const util_1 = require("util");
 exports.execAsync = (0, util_1.promisify)(child_process_1.exec);
@@ -65,37 +69,228 @@ async function isPortInUse(port) {
     }
 }
 /**
- * Get memory usage for a process in bytes
- * Uses 'top' on macOS which includes GPU/Metal memory (more accurate for llama-server)
+ * Spawn a streaming command, read one line, and kill it
+ * Useful for commands like 'macmon pipe' that stream indefinitely
+ * Ensures the process is killed to prevent leaks
+ */
+async function spawnAndReadOneLine(command, args, timeoutMs = 2000) {
+    return new Promise((resolve) => {
+        const child = (0, child_process_1.spawn)(command, args, {
+            stdio: ['ignore', 'pipe', 'ignore'],
+            detached: false, // Keep in same process group for easier cleanup
+        });
+        let resolved = false;
+        let output = '';
+        const cleanup = () => {
+            try {
+                // Try SIGKILL immediately (SIGTERM may not work for macmon)
+                child.kill('SIGKILL');
+            }
+            catch {
+                // Process might already be dead
+            }
+        };
+        // Set timeout to kill process if it doesn't produce output
+        const timeout = setTimeout(() => {
+            if (!resolved) {
+                resolved = true;
+                cleanup();
+                resolve(null);
+            }
+        }, timeoutMs);
+        // Read stdout line by line
+        child.stdout?.on('data', (data) => {
+            if (resolved)
+                return;
+            output += data.toString();
+            // Check if we have a complete line
+            const newlineIndex = output.indexOf('\n');
+            if (newlineIndex !== -1) {
+                const line = output.substring(0, newlineIndex).trim();
+                if (line.length > 0) {
+                    resolved = true;
+                    clearTimeout(timeout);
+                    cleanup();
+                    resolve(line);
+                }
+            }
+        });
+        // Handle process errors
+        child.on('error', () => {
+            if (!resolved) {
+                resolved = true;
+                clearTimeout(timeout);
+                resolve(null);
+            }
+        });
+        // Handle process exit
+        child.on('exit', () => {
+            if (!resolved) {
+                resolved = true;
+                clearTimeout(timeout);
+                // Return partial output if we have any
+                const line = output.trim();
+                resolve(line.length > 0 ? line : null);
+            }
+        });
+    });
+}
+// Process memory cache to prevent spawning too many 'top' processes
+// Cache per PID with 3-second TTL
+const processMemoryCache = new Map();
+const PROCESS_MEMORY_CACHE_TTL = 3000; // 3 seconds
+/**
+ * Batch get memory usage for multiple processes in one top call
+ * Much more efficient than calling getProcessMemory() multiple times
+ * Returns Map<pid, bytes> for all requested PIDs
+ */
+async function getBatchProcessMemory(pids) {
+    const result = new Map();
+    const now = Date.now();
+    // Check cache and collect PIDs that need fetching
+    const pidsToFetch = [];
+    for (const pid of pids) {
+        const cached = processMemoryCache.get(pid);
+        if (cached && (now - cached.timestamp) < PROCESS_MEMORY_CACHE_TTL) {
+            result.set(pid, cached.value);
+        }
+        else {
+            pidsToFetch.push(pid);
+        }
+    }
+    // If all PIDs were cached, return early
+    if (pidsToFetch.length === 0) {
+        return result;
+    }
+    try {
+        // Build top command with all PIDs: top -l 1 -pid X -pid Y -pid Z -stats pid,mem
+        const pidArgs = pidsToFetch.map(pid => `-pid ${pid}`).join(' ');
+        const output = await execCommand(`top -l 1 ${pidArgs} -stats pid,mem 2>/dev/null`);
+        // Parse output: each line is "PID  MEM" (e.g., "1438  299M")
+        const lines = output.split('\n');
+        for (const line of lines) {
+            const match = line.trim().match(/^(\d+)\s+([\d.]+)([KMGT])\s*$/);
+            if (!match)
+                continue;
+            const pid = parseInt(match[1], 10);
+            const value = parseFloat(match[2]);
+            const unit = match[3];
+            // Convert to bytes
+            const multipliers = {
+                K: 1024,
+                M: 1024 * 1024,
+                G: 1024 * 1024 * 1024,
+                T: 1024 * 1024 * 1024 * 1024,
+            };
+            const bytes = Math.round(value * multipliers[unit]);
+            // Cache and store result
+            processMemoryCache.set(pid, { value: bytes, timestamp: now });
+            result.set(pid, bytes);
+        }
+        // For any PIDs that weren't in the output, cache null
+        for (const pid of pidsToFetch) {
+            if (!result.has(pid)) {
+                processMemoryCache.set(pid, { value: null, timestamp: now });
+                result.set(pid, null);
+            }
+        }
+        return result;
+    }
+    catch {
+        // On error, cache null for all requested PIDs
+        for (const pid of pidsToFetch) {
+            processMemoryCache.set(pid, { value: null, timestamp: now });
+            result.set(pid, null);
+        }
+        return result;
+    }
+}
+/**
+ * Get memory usage for a single process in bytes
+ * Uses 'top' on macOS which reports CPU memory only (NOT GPU/Metal memory)
  * Returns null if process not found or error occurs
+ * Caches results for 3 seconds to prevent spawning too many top processes
+ *
+ * Note: For llama-server processes with GPU offloading, use ServerConfig.metalMemoryMB
+ * to get GPU memory allocation (parsed from logs during server startup)
+ *
+ * Note: For multiple PIDs, use getBatchProcessMemory() instead - much more efficient
  */
 async function getProcessMemory(pid) {
+    const result = await getBatchProcessMemory([pid]);
+    return result.get(pid) ?? null;
+}
+// Process CPU cache to prevent spawning too many 'ps' processes
+// Cache per PID with 3-second TTL
+const processCpuCache = new Map();
+const PROCESS_CPU_CACHE_TTL = 3000; // 3 seconds
+/**
+ * Batch get CPU usage for multiple processes in one ps call
+ * Much more efficient than calling getProcessCpu() multiple times
+ * Returns Map<pid, percentage> for all requested PIDs
+ */
+async function getBatchProcessCpu(pids) {
+    const result = new Map();
+    const now = Date.now();
+    // Check cache and collect PIDs that need fetching
+    const pidsToFetch = [];
+    for (const pid of pids) {
+        const cached = processCpuCache.get(pid);
+        if (cached && (now - cached.timestamp) < PROCESS_CPU_CACHE_TTL) {
+            result.set(pid, cached.value);
+        }
+        else {
+            pidsToFetch.push(pid);
+        }
+    }
+    // If all PIDs were cached, return early
+    if (pidsToFetch.length === 0) {
+        return result;
+    }
     try {
-        // Use top with -l 1 (one sample) to get memory stats
-        // MEM column shows resident memory including GPU memory on macOS
-        const output = await execCommand(`top -l 1 -pid ${pid} -stats mem`);
-        // Get the last non-empty line which contains the memory value
-        const lines = output.split('\n').filter((line) => line.trim().length > 0);
-        if (lines.length === 0)
-            return null;
-        const memStr = lines[lines.length - 1].trim();
-        // Parse memory string (e.g., "10.5G", "512M", "1024K", "10G")
-        const match = memStr.match(/^([\d.]+)([KMGT])$/);
-        if (!match)
-            return null;
-        const value = parseFloat(match[1]);
-        const unit = match[2];
-        // Convert to bytes
-        const multipliers = {
-            K: 1024,
-            M: 1024 * 1024,
-            G: 1024 * 1024 * 1024,
-            T: 1024 * 1024 * 1024 * 1024,
-        };
-        return Math.round(value * multipliers[unit]);
+        // Build ps command with all PIDs: ps -p X,Y,Z -o pid=,%cpu=
+        const pidList = pidsToFetch.join(',');
+        const output = await execCommand(`ps -p ${pidList} -o pid=,%cpu= 2>/dev/null`);
+        // Parse output: each line is "PID  %CPU" (e.g., "1438  45.2")
+        const lines = output.split('\n');
+        for (const line of lines) {
+            const match = line.trim().match(/^(\d+)\s+([\d.]+)\s*$/);
+            if (!match)
+                continue;
+            const pid = parseInt(match[1], 10);
+            const cpuPercent = parseFloat(match[2]);
+            // Cache and store result
+            processCpuCache.set(pid, { value: cpuPercent, timestamp: now });
+            result.set(pid, cpuPercent);
+        }
+        // For any PIDs that weren't in the output, cache null (process not running)
+        for (const pid of pidsToFetch) {
+            if (!result.has(pid)) {
+                processCpuCache.set(pid, { value: null, timestamp: now });
+                result.set(pid, null);
+            }
+        }
+        return result;
     }
     catch {
-        return null;
+        // On error, cache null for all requested PIDs
+        for (const pid of pidsToFetch) {
+            processCpuCache.set(pid, { value: null, timestamp: now });
+            result.set(pid, null);
+        }
+        return result;
     }
 }
+/**
+ * Get CPU usage for a single process as percentage (0-100+)
+ * Uses 'ps -o %cpu' on macOS
+ * Returns null if process not found or error occurs
+ * Caches results for 3 seconds to prevent spawning too many ps processes
+ *
+ * Note: For multiple PIDs, use getBatchProcessCpu() instead - much more efficient
+ */
+async function getProcessCpu(pid) {
+    const result = await getBatchProcessCpu([pid]);
+    return result.get(pid) ?? null;
+}
 //# sourceMappingURL=process-utils.js.map

package/dist/utils/process-utils.js.map

@@ -1 +1 @@
-{"version":3,"file":"process-utils.js","sourceRoot":"","sources":["../../src/utils/process-utils.ts"],"names":[],"mappings":";;;AASA,kCAGC;AAKD,0CAMC;AAKD,sCAOC;AAKD,4CAOC;AAKD,kCAOC;AAOD,4CA+BC;AAjGD,iDAAqC;AACrC,+BAAiC;AAEpB,QAAA,SAAS,GAAG,IAAA,gBAAS,EAAC,oBAAI,CAAC,CAAC;AAEzC;;;GAGG;AACI,KAAK,UAAU,WAAW,CAAC,OAAe;IAC/C,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,IAAA,iBAAS,EAAC,OAAO,CAAC,CAAC;IAC5C,OAAO,MAAM,CAAC,IAAI,EAAE,CAAC;AACvB,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,eAAe,CAAC,OAAe;IACnD,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,IAAA,iBAAS,EAAC,OAAO,CAAC,CAAC;IACpD,OAAO;QACL,MAAM,EAAE,MAAM,CAAC,IAAI,EAAE;QACrB,MAAM,EAAE,MAAM,CAAC,IAAI,EAAE;KACtB,CAAC;AACJ,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,aAAa,CAAC,OAAe;IACjD,IAAI,CAAC;QACH,MAAM,IAAA,iBAAS,EAAC,SAAS,OAAO,EAAE,CAAC,CAAC;QACpC,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,gBAAgB,CAAC,GAAW;IAChD,IAAI,CAAC;QACH,MAAM,IAAA,iBAAS,EAAC,SAAS,GAAG,EAAE,CAAC,CAAC;QAChC,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,WAAW,CAAC,IAAY;IAC5C,IAAI,CAAC;QACH,MAAM,IAAA,iBAAS,EAAC,cAAc,IAAI,kBAAkB,CAAC,CAAC;QACtD,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;;;GAIG;AACI,KAAK,UAAU,gBAAgB,CAAC,GAAW;IAChD,IAAI,CAAC;QACH,qDAAqD;QACrD,iEAAiE;QACjE,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,iBAAiB,GAAG,aAAa,CAAC,CAAC;QAEpE,8DAA8D;QAC9D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC1E,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QAEpC,MAAM,MAAM,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAE9C,8DAA8D;QAC9D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,oBAAoB,CAAC,CAAC;QACjD,IAAI,CAAC,KAAK;YAAE,OAAO,IAAI,CAAC;QAExB,MAAM,KAAK,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;QACnC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QAEtB,mBAAmB;QACnB,MAAM,WAAW,GAA8B;YAC7C,CAAC,EAAE,IAAI;YACP,CAAC,EAAE,IAAI,GAAG,IAAI;YACd,CAAC,EAAE,IAAI,GAAG,IAAI,GAAG,IAAI;YACrB,CAAC,EAAE,IAAI,GAAG,IAAI,GAAG,IAAI,GAAG,IAAI;SAC7B,CAAC;QAEF,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC;IAC/C,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC"}
+{"version":3,"file":"process-utils.js","sourceRoot":"","sources":["../../src/utils/process-utils.ts"],"names":[],"mappings":";;;AASA,kCAGC;AAKD,0CAMC;AAKD,sCAOC;AAKD,4CAOC;AAKD,kCAOC;AAOD,kDAyEC;AAYD,sDAmEC;AAaD,4CAGC;AAYD,gDAwDC;AAUD,sCAGC;AA3TD,iDAA4C;AAC5C,+BAAiC;AAEpB,QAAA,SAAS,GAAG,IAAA,gBAAS,EAAC,oBAAI,CAAC,CAAC;AAEzC;;;GAGG;AACI,KAAK,UAAU,WAAW,CAAC,OAAe;IAC/C,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,IAAA,iBAAS,EAAC,OAAO,CAAC,CAAC;IAC5C,OAAO,MAAM,CAAC,IAAI,EAAE,CAAC;AACvB,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,eAAe,CAAC,OAAe;IACnD,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,IAAA,iBAAS,EAAC,OAAO,CAAC,CAAC;IACpD,OAAO;QACL,MAAM,EAAE,MAAM,CAAC,IAAI,EAAE;QACrB,MAAM,EAAE,MAAM,CAAC,IAAI,EAAE;KACtB,CAAC;AACJ,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,aAAa,CAAC,OAAe;IACjD,IAAI,CAAC;QACH,MAAM,IAAA,iBAAS,EAAC,SAAS,OAAO,EAAE,CAAC,CAAC;QACpC,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,gBAAgB,CAAC,GAAW;IAChD,IAAI,CAAC;QACH,MAAM,IAAA,iBAAS,EAAC,SAAS,GAAG,EAAE,CAAC,CAAC;QAChC,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;GAEG;AACI,KAAK,UAAU,WAAW,CAAC,IAAY;IAC5C,IAAI,CAAC;QACH,MAAM,IAAA,iBAAS,EAAC,cAAc,IAAI,kBAAkB,CAAC,CAAC;QACtD,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;;;GAIG;AACI,KAAK,UAAU,mBAAmB,CACvC,OAAe,EACf,IAAc,EACd,YAAoB,IAAI;IAExB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE;QAC7B,MAAM,KAAK,GAAG,IAAA,qBAAK,EAAC,OAAO,EAAE,IAAI,EAAE;YACjC,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC;YACnC,QAAQ,EAAE,KAAK,EAAE,gDAAgD;SAClE,CAAC,CAAC;QAEH,IAAI,QAAQ,GAAG,KAAK,CAAC;QACrB,IAAI,MAAM,GAAG,EAAE,CAAC;QAEhB,MAAM,OAAO,GAAG,GAAG,EAAE;YACnB,IAAI,CAAC;gBACH,4DAA4D;gBAC5D,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YACxB,CAAC;YAAC,MAAM,CAAC;gBACP,gCAAgC;YAClC,CAAC;QACH,CAAC,CAAC;QAEF,2DAA2D;QAC3D,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE;YAC9B,IAAI,CAAC,QAAQ,EAAE,CAAC;gBACd,QAAQ,GAAG,IAAI,CAAC;gBAChB,OAAO,EAAE,CAAC;gBACV,OAAO,CAAC,IAAI,CAAC,CAAC;YAChB,CAAC;QACH,CAAC,EAAE,SAAS,CAAC,CAAC;QAEd,2BAA2B;QAC3B,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE;YAChC,IAAI,QAAQ;gBAAE,OAAO;YAErB,MAAM,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAE1B,mCAAmC;YACnC,MAAM,YAAY,GAAG,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YAC1C,IAAI,YAAY,KAAK,CAAC,CAAC,EAAE,CAAC;gBACxB,MAAM,IAAI,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,EAAE,YAAY,CAAC,CAAC,IAAI,EAAE,CAAC;gBAEtD,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;oBACpB,QAAQ,GAAG,IAAI,CAAC;oBAChB,YAAY,CAAC,OAAO,CAAC,CAAC;oBACtB,OAAO,EAAE,CAAC;oBACV,OAAO,CAAC,IAAI,CAAC,CAAC;gBAChB,CAAC;YACH,CAAC;QACH,CAAC,CAAC,CAAC;QAEH,wBAAwB;QACxB,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE;YACrB,IAAI,CAAC,QAAQ,EAAE,CAAC;gBACd,QAAQ,GAAG,IAAI,CAAC;gBAChB,YAAY,CAAC,OAAO,CAAC,CAAC;gBACtB,OAAO,CAAC,IAAI,CAAC,CAAC;YAChB,CAAC;QACH,CAAC,CAAC,CAAC;QAEH,sBAAsB;QACtB,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,GAAG,EAAE;YACpB,IAAI,CAAC,QAAQ,EAAE,CAAC;gBACd,QAAQ,GAAG,IAAI,CAAC;gBAChB,YAAY,CAAC,OAAO,CAAC,CAAC;gBAEtB,uCAAuC;gBACvC,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;gBAC3B,OAAO,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;YACzC,CAAC;QACH,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED,oEAAoE;AACpE,kCAAkC;AAClC,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAuD,CAAC;AAC1F,MAAM,wBAAwB,GAAG,IAAI,CAAC,CAAC,YAAY;AAEnD;;;;GAIG;AACI,KAAK,UAAU,qBAAqB,CAAC,IAAc;IACxD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAyB,CAAC;IAChD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAEvB,kDAAkD;IAClD,MAAM,WAAW,GAAa,EAAE,CAAC;IACjC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,MAAM,GAAG,kBAAkB,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC3C,IAAI,MAAM,IAAI,CAAC,GAAG,GAAG,MAAM,CAAC,SAAS,CAAC,GAAG,wBAAwB,EAAE,CAAC;YAClE,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAChC,CAAC;aAAM,CAAC;YACN,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;IAED,wCAAwC;IACxC,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7B,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,IAAI,CAAC;QACH,gFAAgF;QAChF,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAChE,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,YAAY,OAAO,6BAA6B,CAAC,CAAC;QAEnF,6DAA6D;QAC7D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,+BAA+B,CAAC,CAAC;YACjE,IAAI,CAAC,KAAK;gBAAE,SAAS;YAErB,MAAM,GAAG,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACnC,MAAM,KAAK,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;YACnC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;YAEtB,mBAAmB;YACnB,MAAM,WAAW,GAA8B;gBAC7C,CAAC,EAAE,IAAI;gBACP,CAAC,EAAE,IAAI,GAAG,IAAI;gBACd,CAAC,EAAE,IAAI,GAAG,IAAI,GAAG,IAAI;gBACrB,CAAC,EAAE,IAAI,GAAG,IAAI,GAAG,IAAI,GAAG,IAAI;aAC7B,CAAC;YAEF,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC;YAEpD,yBAAyB;YACzB,kBAAkB,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;YAC9D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACzB,CAAC;QAED,sDAAsD;QACtD,KAAK,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YAC9B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;gBACrB,kBAAkB,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;gBAC7D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;YACxB,CAAC;QACH,CAAC;QAED,OAAO,MAAM,CAAC;IAChB,CAAC;IAAC,MAAM,CAAC;QACP,8CAA8C;QAC9C,KAAK,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YAC9B,kBAAkB,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;YAC7D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QACxB,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;AACH,CAAC;AAED;;;;;;;;;;GAUG;AACI,KAAK,UAAU,gBAAgB,CAAC,GAAW;IAChD,MAAM,MAAM,GAAG,MAAM,qBAAqB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAClD,OAAO,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC;AACjC,CAAC;AAED,gEAAgE;AAChE,kCAAkC;AAClC,MAAM,eAAe,GAAG,IAAI,GAAG,EAAuD,CAAC;AACvF,MAAM,qBAAqB,GAAG,IAAI,CAAC,CAAC,YAAY;AAEhD;;;;GAIG;AACI,KAAK,UAAU,kBAAkB,CAAC,IAAc;IACrD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAyB,CAAC;IAChD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAEvB,kDAAkD;IAClD,MAAM,WAAW,GAAa,EAAE,CAAC;IACjC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACxC,IAAI,MAAM,IAAI,CAAC,GAAG,GAAG,MAAM,CAAC,SAAS,CAAC,GAAG,qBAAqB,EAAE,CAAC;YAC/D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAChC,CAAC;aAAM,CAAC;YACN,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;IAED,wCAAwC;IACxC,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7B,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,IAAI,CAAC;QACH,4DAA4D;QAC5D,MAAM,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACtC,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,SAAS,OAAO,4BAA4B,CAAC,CAAC;QAE/E,8DAA8D;QAC9D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAC;YACzD,IAAI,CAAC,KAAK;gBAAE,SAAS;YAErB,MAAM,GAAG,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACnC,MAAM,UAAU,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;YAExC,yBAAyB;YACzB,eAAe,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;YAChE,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;QAC9B,CAAC;QAED,4EAA4E;QAC5E,KAAK,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YAC9B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;gBACrB,eAAe,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;gBAC1D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;YACxB,CAAC;QACH,CAAC;QAED,OAAO,MAAM,CAAC;IAChB,CAAC;IAAC,MAAM,CAAC;QACP,8CAA8C;QAC9C,KAAK,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YAC9B,eAAe,CAAC,GAAG,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;YAC1D,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QACxB,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;AACH,CAAC;AAED;;;;;;;GAOG;AACI,KAAK,UAAU,aAAa,CAAC,GAAW;IAC7C,MAAM,MAAM,GAAG,MAAM,kBAAkB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/C,OAAO,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC;AACjC,CAAC"}

package/docs/images/.gitkeep

@@ -0,0 +1 @@
+# Screenshots directory

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@appkit/llamacpp-cli",
-  "version": "1.4.1",
+  "version": "1.6.0",
   "description": "CLI tool to manage local llama.cpp servers on macOS",
   "main": "dist/cli.js",
   "bin": {
@@ -40,11 +40,15 @@
     "url": "https://github.com/appkitstudio/llamacpp-cli/issues"
   },
   "dependencies": {
+    "asciichart": "^1.5.25",
+    "blessed": "^0.1.81",
     "chalk": "^4.1.2",
     "cli-table3": "^0.6.5",
     "commander": "^13.0.0"
   },
   "devDependencies": {
+    "@types/asciichart": "^1.5.8",
+    "@types/blessed": "^0.1.27",
     "@types/node": "^20.12.7",
     "commit-and-tag-version": "^12.6.1",
     "tsx": "^4.7.2",

package/src/cli.ts

@@ -17,6 +17,7 @@ import { showCommand } from './commands/show';
 import { serverShowCommand } from './commands/server-show';
 import { serverConfigCommand } from './commands/config';
 import { configGlobalCommand } from './commands/config-global';
+import { monitorCommand } from './commands/monitor';
 import packageJson from '../package.json';
 
 const program = new Command();
@@ -41,11 +42,12 @@ program
 
 // List running servers
 program
-  .command('ps')
-  .description('List all servers with status')
-  .action(async () => {
+  .command('ps [identifier]')
+  .description('Interactive server monitoring dashboard')
+  .option('--table', 'Show static table instead of TUI (for scripting)')
+  .action(async (identifier?: string, options?: { table?: boolean }) => {
     try {
-      await psCommand();
+      await psCommand(identifier, options);
     } catch (error) {
       console.error(chalk.red('❌ Error:'), (error as Error).message);
       process.exit(1);
@@ -268,5 +270,20 @@ server
     }
   });
 
+// Monitor server (deprecated - redirects to ps)
+server
+  .command('monitor [identifier]')
+  .description('Monitor server with real-time metrics TUI (deprecated: use "llamacpp ps" instead)')
+  .action(async (identifier?: string) => {
+    try {
+      console.log(chalk.yellow('⚠️  The "monitor" command is deprecated and will be removed in a future version.'));
+      console.log(chalk.dim('   Please use "llamacpp ps" instead for the same functionality.\n'));
+      await monitorCommand(identifier);
+    } catch (error) {
+      console.error(chalk.red('❌ Error:'), (error as Error).message);
+      process.exit(1);
+    }
+  });
+
 // Parse arguments
 program.parse();

package/src/commands/create.ts

@@ -9,7 +9,7 @@ import { launchctlManager } from '../lib/launchctl-manager';
 import { statusChecker } from '../lib/status-checker';
 import { commandExists } from '../utils/process-utils';
 import { formatBytes } from '../utils/format-utils';
-import { ensureDir } from '../utils/file-utils';
+import { ensureDir, parseMetalMemoryFromLog } from '../utils/file-utils';
 import { ensureModelsDirectory } from '../lib/models-dir-setup';
 
 interface CreateOptions {
@@ -159,12 +159,22 @@ export async function createCommand(model: string, options: CreateOptions): Prom
   }
 
   // 12. Update config with running status
-  const updatedConfig = await statusChecker.updateServerStatus(config);
+  let updatedConfig = await statusChecker.updateServerStatus(config);
+
+  // 13. Parse Metal (GPU) memory allocation from logs
+  // Wait a few seconds for model to start loading (large models take time)
+  console.log(chalk.dim('Detecting Metal (GPU) memory allocation...'));
+  await new Promise(resolve => setTimeout(resolve, 8000)); // 8 second delay
+  const metalMemoryMB = await parseMetalMemoryFromLog(updatedConfig.stderrPath);
+  if (metalMemoryMB) {
+    updatedConfig = { ...updatedConfig, metalMemoryMB };
+    console.log(chalk.dim(`Metal memory: ${metalMemoryMB.toFixed(0)} MB`));
+  }
 
-  // 13. Save server config
+  // 14. Save server config
   await stateManager.saveServerConfig(updatedConfig);
 
-  // 14. Display success message
+  // 15. Display success message
   console.log();
   console.log(chalk.green('✅ Server created and started successfully!'));
   console.log();

package/src/commands/monitor.ts

@@ -0,0 +1,110 @@
+import chalk from 'chalk';
+import blessed from 'blessed';
+import { stateManager } from '../lib/state-manager.js';
+import { createMonitorUI } from '../tui/MonitorApp.js';
+import { createMultiServerMonitorUI } from '../tui/MultiServerMonitorApp.js';
+
+export async function monitorCommand(identifier?: string): Promise<void> {
+  // Initialize state manager
+  await stateManager.initialize();
+
+  // Get all servers
+  const allServers = await stateManager.getAllServers();
+  if (allServers.length === 0) {
+    throw new Error(
+      `No servers configured.\n\n` +
+        `Create a server first: llamacpp server create <model>`
+    );
+  }
+
+  // Create blessed screen
+  const screen = blessed.screen({
+    smartCSR: true,
+    title: 'llama.cpp Server Monitor',
+  });
+
+  // Determine which UI to launch
+  if (identifier) {
+    // User specified a server - single server mode
+    const server = await stateManager.findServer(identifier);
+    if (!server) {
+      screen.destroy();
+      throw new Error(
+        `Server not found: ${identifier}\n\n` +
+          `Use: llamacpp ps\n` +
+          `Or create a new server: llamacpp server create <model>`
+      );
+    }
+
+    // Update server status to reflect actual launchctl state
+    const { statusChecker } = await import('../lib/status-checker.js');
+    const status = await statusChecker.checkServer(server);
+    server.status = status.isRunning ? 'running' : 'stopped';
+    server.pid = status.pid || undefined;
+
+    // Check if server is running
+    if (server.status !== 'running') {
+      screen.destroy();
+      throw new Error(
+        `Server ${server.modelName} is not running.\n\n` +
+          `Start it first: llamacpp server start ${server.id}`
+      );
+    }
+
+    // Launch single-server TUI
+    await createMonitorUI(screen, server);
+  } else if (allServers.length === 1) {
+    // Only one server - single server mode
+    const server = allServers[0];
+
+    // Update server status to reflect actual launchctl state
+    const { statusChecker } = await import('../lib/status-checker.js');
+    const status = await statusChecker.checkServer(server);
+    server.status = status.isRunning ? 'running' : 'stopped';
+    server.pid = status.pid || undefined;
+
+    // Check if server is running
+    if (server.status !== 'running') {
+      screen.destroy();
+      throw new Error(
+        `Server ${server.modelName} is not running.\n\n` +
+          `Start it first: llamacpp server start ${server.id}`
+      );
+    }
+
+    // Launch single-server TUI
+    await createMonitorUI(screen, server);
+  } else {
+    // Multiple servers - multi-server mode
+    // Update all server statuses to reflect actual launchctl state
+    const { statusChecker } = await import('../lib/status-checker.js');
+    for (const server of allServers) {
+      const status = await statusChecker.checkServer(server);
+      server.status = status.isRunning ? 'running' : 'stopped';
+      server.pid = status.pid || undefined;
+    }
+
+    // Filter to only running servers for monitoring
+    const runningServers = allServers.filter(s => s.status === 'running');
+
+    if (runningServers.length === 0) {
+      screen.destroy();
+      throw new Error(
+        `No servers are currently running.\n\n` +
+          `Start a server first:\n` +
+          allServers
+            .map((s) => `  llamacpp server start ${s.id}  # ${s.modelName}`)
+            .join('\n')
+      );
+    }
+
+    // Launch multi-server TUI (pass all servers so we can see stopped ones too)
+    await createMultiServerMonitorUI(screen, allServers);
+  }
+
+  // Render the screen
+  screen.render();
+
+  // Note: TUI functions handle their own key events and exit directly
+  // The process will stay alive until user presses Q/Ctrl+C
+}

package/src/commands/ps.ts

@@ -1,11 +1,14 @@
 import chalk from 'chalk';
 import Table from 'cli-table3';
+import blessed from 'blessed';
 import { stateManager } from '../lib/state-manager';
 import { statusChecker } from '../lib/status-checker';
 import { formatUptime, formatBytes } from '../utils/format-utils';
 import { getProcessMemory } from '../utils/process-utils';
+import { createMultiServerMonitorUI } from '../tui/MultiServerMonitorApp.js';
+import { ServerConfig } from '../types/server-config.js';
 
-export async function psCommand(): Promise<void> {
+async function showStaticTable(): Promise<void> {
   const servers = await stateManager.getAllServers();
 
   if (servers.length === 0) {
@@ -52,12 +55,14 @@ export async function psCommand(): Promise<void> {
         ? formatUptime(server.lastStarted)
         : '-';
 
-    // Get memory usage for running servers
+    // Get memory usage for running servers (CPU + Metal GPU memory)
     let memoryText = '-';
     if (server.status === 'running' && server.pid) {
-      const memoryBytes = await getProcessMemory(server.pid);
-      if (memoryBytes !== null) {
-        memoryText = formatBytes(memoryBytes);
+      const cpuMemoryBytes = await getProcessMemory(server.pid);
+      if (cpuMemoryBytes !== null) {
+        const metalMemoryBytes = server.metalMemoryMB ? server.metalMemoryMB * 1024 * 1024 : 0;
+        const totalMemoryBytes = cpuMemoryBytes + metalMemoryBytes;
+        memoryText = formatBytes(totalMemoryBytes);
       }
     }
 
@@ -88,3 +93,81 @@ export async function psCommand(): Promise<void> {
     console.log(chalk.red('\n⚠️  Some servers have crashed. Check logs with: llamacpp server logs <id> --errors'));
   }
 }
+
+export async function psCommand(identifier?: string, options?: { table?: boolean }): Promise<void> {
+  // If --table flag is set, show static table (backward compatibility)
+  if (options?.table) {
+    await showStaticTable();
+    return;
+  }
+
+  // Get all servers and update their statuses
+  const servers = await stateManager.getAllServers();
+
+  if (servers.length === 0) {
+    console.log(chalk.yellow('No servers configured.'));
+    console.log(chalk.dim('\nCreate a server: llamacpp server create <model-filename>'));
+    return;
+  }
+
+  // Update all server statuses
+  const updated = await statusChecker.updateAllServerStatuses();
+
+  // If identifier is provided, find the server and jump to detail view
+  if (identifier) {
+    const server = await findServer(identifier, updated);
+    if (!server) {
+      console.log(chalk.red(`❌ Server not found: ${identifier}`));
+      console.log(chalk.dim('\nAvailable servers:'));
+      updated.forEach((s: ServerConfig) => {
+        console.log(chalk.dim(`  - ${s.id} (port ${s.port})`));
+      });
+      process.exit(1);
+    }
+
+    // Find the server index for direct jump
+    const serverIndex = updated.findIndex(s => s.id === server.id);
+
+    // Launch multi-server TUI with direct jump to detail view
+    const screen = blessed.screen({
+      smartCSR: true,
+      title: 'llama.cpp Multi-Server Monitor',
+      fullUnicode: true,
+    });
+
+    await createMultiServerMonitorUI(screen, updated, true, serverIndex); // fromPs = true, directJumpIndex
+    return;
+  }
+
+  // No identifier - launch multi-server TUI
+  const runningServers = updated.filter((s: ServerConfig) => s.status === 'running');
+
+  // Launch multi-server TUI (shows all servers, not just running ones)
+  const screen = blessed.screen({
+    smartCSR: true,
+    title: 'llama.cpp Multi-Server Monitor',
+    fullUnicode: true,
+  });
+
+  await createMultiServerMonitorUI(screen, updated, true); // fromPs = true
+}
+
+// Helper function to find server by identifier
+async function findServer(identifier: string, servers: ServerConfig[]): Promise<ServerConfig | null> {
+  // Try by port
+  const port = parseInt(identifier);
+  if (!isNaN(port)) {
+    const server = servers.find(s => s.port === port);
+    if (server) return server;
+  }
+
+  // Try by exact ID
+  const byId = servers.find(s => s.id === identifier);
+  if (byId) return byId;
+
+  // Try by partial model name
+  const byModel = servers.find(s => s.modelName.toLowerCase().includes(identifier.toLowerCase()));
+  if (byModel) return byModel;
+
+  return null;
+}

package/src/commands/server-show.ts

@@ -68,9 +68,16 @@ export async function serverShowCommand(identifier: string): Promise<void> {
     }
 
     if (updatedServer.pid) {
-      const memoryBytes = await getProcessMemory(updatedServer.pid);
-      if (memoryBytes !== null) {
-        console.log(`${chalk.bold('Memory:')}         ${formatBytes(memoryBytes)}`);
+      const cpuMemoryBytes = await getProcessMemory(updatedServer.pid);
+      if (cpuMemoryBytes !== null) {
+        const metalMemoryBytes = updatedServer.metalMemoryMB ? updatedServer.metalMemoryMB * 1024 * 1024 : 0;
+        const totalMemoryBytes = cpuMemoryBytes + metalMemoryBytes;
+
+        if (metalMemoryBytes > 0) {
+          console.log(`${chalk.bold('Memory:')}         ${formatBytes(totalMemoryBytes)} (CPU: ${formatBytes(cpuMemoryBytes)}, GPU: ${formatBytes(metalMemoryBytes)})`);
+        } else {
+          console.log(`${chalk.bold('Memory:')}         ${formatBytes(cpuMemoryBytes)} (CPU only)`);
+        }
       }
     }
   }