quilltap 4.4.0-dev.99 → 4.5.0-dev

package/bin/quilltap.js

@@ -169,6 +169,19 @@ function ensureNativeModules() {
     }
   }
 
+  // Check node-pty: backs the Ariel terminal feature. Loaded dynamically by
+  // pty-manager in the standalone server, so resolution must succeed and the
+  // native binding's NODE_MODULE_VERSION must match the runtime.
+  try {
+    require('node-pty');
+  } catch (err) {
+    if (err.message && err.message.includes('NODE_MODULE_VERSION')) {
+      needsRebuild.push('node-pty');
+    } else if (err.code === 'MODULE_NOT_FOUND') {
+      needsRebuild.push('node-pty');
+    }
+  }
+
   if (needsRebuild.length === 0) return;
 
   console.log(`  Rebuilding native modules for Node.js ${process.version}...`);
@@ -238,6 +251,28 @@ function linkNativeModules(standaloneDir) {
                         || resolveModuleDir('better-sqlite3');
   linkModule('better-sqlite3', betterSqlite3Dir);
 
+  // Link node-pty — the standalone tarball strips it (platform-specific),
+  // and pty-manager loads it via a dynamic require, so it needs to resolve
+  // from standaloneDir/node_modules.
+  const nodePtyDir = resolveModuleDir('node-pty');
+  linkModule('node-pty', nodePtyDir);
+  if (nodePtyDir) {
+    // Some npm cache extractions strip the executable bit on spawn-helper,
+    // causing pty.spawn() to fail with `posix_spawnp failed`. Restore it.
+    const prebuildsDir = path.join(nodePtyDir, 'prebuilds');
+    if (fs.existsSync(prebuildsDir)) {
+      try {
+        for (const entry of fs.readdirSync(prebuildsDir, { withFileTypes: true })) {
+          if (!entry.isDirectory()) continue;
+          const helper = path.join(prebuildsDir, entry.name, 'spawn-helper');
+          if (fs.existsSync(helper)) {
+            try { fs.chmodSync(helper, 0o755); } catch { /* best-effort */ }
+          }
+        }
+      } catch { /* best-effort */ }
+    }
+  }
+
   // Link sharp
   const sharpDir = resolveModuleDir('sharp');
   linkModule('sharp', sharpDir);

package/lib/memory-diff-command.js

@@ -38,16 +38,20 @@ Nothing is persisted; the server runs the extraction passes in dry-run mode
 so the comparison is non-destructive.
 
 Options:
-  -d, --data-dir <path>     Override data directory (instance root)
-      --passphrase <pass>   Decrypt .dbkey if peppered
-      --port <number>       Server port for API calls (default: 3000)
-      --out <dir>           Output directory (default: cwd)
-  -h, --help                Show this help
+  -d, --data-dir <path>      Override data directory (instance root)
+      --passphrase <pass>    Decrypt .dbkey if peppered
+      --port <number>        Server port for API calls (default: 3000)
+      --out <dir>            Output directory (default: cwd)
+      --concurrency <number> Parallel turns to process (default: 4, max: 32).
+                             Cloud cheap-LLMs can handle 8–16; keep this low
+                             for local Ollama to avoid saturating the model.
+  -h, --help                 Show this help
 
 Examples:
   quilltap memory-diff <chatId>
   quilltap memory-diff <chatId> --data-dir ~/iCloud/Quilltap/Friday
   quilltap memory-diff <chatId> --out /tmp/extract-diff
+  quilltap memory-diff <chatId> --concurrency 8     # cloud cheap-LLM
 `);
 }
 
@@ -57,6 +61,7 @@ function parseFlags(args) {
     passphrase: '',
     port: 3000,
     out: process.cwd(),
+    concurrency: 4,
     help: false,
   };
   const positional = [];
@@ -80,6 +85,15 @@ function parseFlags(args) {
         flags.port = p;
         break;
       }
+      case '--concurrency': {
+        const n = parseInt(args[++i], 10);
+        if (isNaN(n) || n < 1 || n > 32) {
+          console.error('Error: --concurrency must be between 1 and 32');
+          process.exit(1);
+        }
+        flags.concurrency = n;
+        break;
+      }
       case '--out':
         flags.out = args[++i];
         break;
@@ -245,8 +259,12 @@ async function memoryDiffCommand(args) {
   process.stderr.write(`  wrote ${GREEN}${existing.length}${RESET} memories to ${existingPath}\n`);
 
   // -------- 2. Stream dry-run re-extraction from the server ----------------
-  const url = `http://localhost:${flags.port}/api/v1/chats/${encodeURIComponent(chatId)}?action=extract-memories-dry-run`;
-  process.stderr.write(`${BOLD}Streaming re-extraction${RESET} from ${DIM}${url}${RESET}\n`);
+  const url =
+    `http://localhost:${flags.port}/api/v1/chats/${encodeURIComponent(chatId)}` +
+    `?action=extract-memories-dry-run&concurrency=${flags.concurrency}`;
+  process.stderr.write(
+    `${BOLD}Streaming re-extraction${RESET} (concurrency ${CYAN}${flags.concurrency}${RESET}) from ${DIM}${url}${RESET}\n`
+  );
 
   let res;
   try {
@@ -299,6 +317,10 @@ async function memoryDiffCommand(args) {
           `  ${RED}[${event.index + 1}/${turnCount}] FAILED${RESET}: ${event.error}\n`
         );
         break;
+      case 'ping':
+        // Server-side heartbeat keeping the connection warm during long
+        // first-turn LLM passes; nothing to display.
+        break;
       case 'done':
         totalCandidates = event.totalCandidates;
         break;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "quilltap",
-  "version": "4.4.0-dev.99",
+  "version": "4.5.0-dev",
   "description": "Self-hosted AI workspace for writers, worldbuilders, and roleplayers. Run with npx quilltap.",
   "author": {
     "name": "Charles Sebold",
@@ -33,12 +33,14 @@
     "README.md"
   ],
   "dependencies": {
+    "@napi-rs/canvas": "^0.1.100",
     "better-sqlite3-multiple-ciphers": "^12.9.0",
+    "node-pty": "^1.1.0",
     "sharp": "^0.34.5",
-    "tar": "^7.5.13",
+    "tar": "^7.5.15",
     "yauzl": "^3.3.0"
   },
   "engines": {
-    "node": ">=22.0.0"
+    "node": ">=24.0.0"
   }
 }