omnius 1.0.104 → 1.0.106

package/dist/index.js

@@ -578020,10 +578020,43 @@ async function needsTextToolMode(modelName, backendUrl2) {
   const hasTools = await checkToolSupport(modelName, backendUrl2);
   return !hasTools;
 }
+function parseNvidiaSmi(stdout) {
+  let total = 0;
+  let free = 0;
+  let name10 = "";
+  const lines = stdout.trim().split("\n").filter(Boolean);
+  for (const line of lines) {
+    const parts = line.split(",").map((s2) => s2.trim());
+    const totMB = parseInt(parts[0] ?? "0", 10);
+    const freeMB = parseInt(parts[1] ?? "0", 10);
+    if (!isNaN(totMB)) total += totMB / 1024;
+    if (!isNaN(freeMB)) free += freeMB / 1024;
+    if (!name10 && parts[2]) name10 = parts[2];
+  }
+  return { total, free, name: name10 };
+}
+function parseRocmSmi(stdout) {
+  let total = 0;
+  let used = 0;
+  let name10 = "";
+  const lines = stdout.trim().split("\n").filter(Boolean);
+  for (const line of lines) {
+    if (line.toLowerCase().startsWith("device") || line.startsWith("=")) continue;
+    const parts = line.split(",").map((s2) => s2.trim());
+    if (parts.length < 3) continue;
+    const usedBytes = parseInt(parts[1] ?? "0", 10);
+    const totBytes = parseInt(parts[2] ?? "0", 10);
+    if (!isNaN(usedBytes)) used += usedBytes / 1024 ** 3;
+    if (!isNaN(totBytes)) total += totBytes / 1024 ** 3;
+    if (!name10 && parts[0]) name10 = parts[0];
+  }
+  return { total, free: Math.max(0, total - used), name: name10 ? `AMD ${name10}` : "AMD GPU" };
+}
 function detectSystemSpecs() {
   let totalRamGB = 0;
   let availableRamGB = 0;
   let gpuVramGB = 0;
+  let availableVramGB = 0;
   let gpuName = "";
   try {
     const memInfo = execSync50("free -b 2>/dev/null || sysctl -n hw.memsize 2>/dev/null", {
@@ -578047,24 +578080,33 @@ function detectSystemSpecs() {
   }
   try {
     const nvidiaSmi = execSync50(
-      "nvidia-smi --query-gpu=memory.total,name --format=csv,noheader,nounits 2>/dev/null",
+      "nvidia-smi --query-gpu=memory.total,memory.free,name --format=csv,noheader,nounits 2>/dev/null",
       { encoding: "utf8", timeout: 5e3 }
     );
-    const lines = nvidiaSmi.trim().split("\n");
-    if (lines.length > 0) {
-      for (const line of lines) {
-        const parts = line.split(",").map((s2) => s2.trim());
-        const vramMB = parseInt(parts[0] ?? "0", 10);
-        if (!isNaN(vramMB)) gpuVramGB += vramMB / 1024;
-        if (!gpuName && parts[1]) gpuName = parts[1];
-      }
-    }
+    const r2 = parseNvidiaSmi(nvidiaSmi);
+    gpuVramGB += r2.total;
+    availableVramGB += r2.free;
+    if (!gpuName && r2.name) gpuName = r2.name;
   } catch {
   }
+  if (gpuVramGB === 0) {
+    try {
+      const rocmSmi = execSync50(
+        "rocm-smi --showmeminfo vram --csv 2>/dev/null",
+        { encoding: "utf8", timeout: 5e3 }
+      );
+      const r2 = parseRocmSmi(rocmSmi);
+      gpuVramGB += r2.total;
+      availableVramGB += r2.free;
+      if (!gpuName && r2.name) gpuName = r2.name;
+    } catch {
+    }
+  }
   return {
     totalRamGB: Math.round(totalRamGB * 10) / 10,
     availableRamGB: Math.round(availableRamGB * 10) / 10,
     gpuVramGB: Math.round(gpuVramGB * 10) / 10,
+    availableVramGB: Math.round(availableVramGB * 10) / 10,
     gpuName
   };
 }
@@ -578072,6 +578114,7 @@ async function detectSystemSpecsAsync() {
   let totalRamGB = 0;
   let availableRamGB = 0;
   let gpuVramGB = 0;
+  let availableVramGB = 0;
   let gpuName = "";
   try {
     const { stdout: memInfo } = await execAsync2(
@@ -578095,24 +578138,33 @@ async function detectSystemSpecsAsync() {
   }
   try {
     const { stdout: nvidiaSmi } = await execAsync2(
-      "nvidia-smi --query-gpu=memory.total,name --format=csv,noheader,nounits 2>/dev/null",
+      "nvidia-smi --query-gpu=memory.total,memory.free,name --format=csv,noheader,nounits 2>/dev/null",
       { timeout: 5e3 }
     );
-    const lines = nvidiaSmi.trim().split("\n");
-    if (lines.length > 0) {
-      for (const line of lines) {
-        const parts = line.split(",").map((s2) => s2.trim());
-        const vramMB = parseInt(parts[0] ?? "0", 10);
-        if (!isNaN(vramMB)) gpuVramGB += vramMB / 1024;
-        if (!gpuName && parts[1]) gpuName = parts[1];
-      }
-    }
+    const r2 = parseNvidiaSmi(nvidiaSmi);
+    gpuVramGB += r2.total;
+    availableVramGB += r2.free;
+    if (!gpuName && r2.name) gpuName = r2.name;
   } catch {
   }
+  if (gpuVramGB === 0) {
+    try {
+      const { stdout: rocmSmi } = await execAsync2(
+        "rocm-smi --showmeminfo vram --csv 2>/dev/null",
+        { timeout: 5e3 }
+      );
+      const r2 = parseRocmSmi(rocmSmi);
+      gpuVramGB += r2.total;
+      availableVramGB += r2.free;
+      if (!gpuName && r2.name) gpuName = r2.name;
+    } catch {
+    }
+  }
   return {
     totalRamGB: Math.round(totalRamGB * 10) / 10,
     availableRamGB: Math.round(availableRamGB * 10) / 10,
     gpuVramGB: Math.round(gpuVramGB * 10) / 10,
+    availableVramGB: Math.round(availableVramGB * 10) / 10,
     gpuName
   };
 }
@@ -578128,7 +578180,9 @@ function recommendModel(specs) {
   return QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:cloud");
 }
 function calculateContextWindow(specs, modelSizeGB2, kvBytesPerToken, archMax) {
-  const totalAvail = Math.max(specs.gpuVramGB, specs.totalRamGB);
+  const ramBudget = specs.availableRamGB > 0 ? specs.availableRamGB : specs.totalRamGB;
+  const vramBudget = specs.availableVramGB > 0 ? specs.availableVramGB : specs.gpuVramGB;
+  const totalAvail = Math.max(vramBudget, ramBudget);
   const remaining = Math.max(0, totalAvail - modelSizeGB2);
   const usableGB = remaining * 0.85;
   let numCtx;
@@ -578155,11 +578209,18 @@ function formatContextLabel(numCtx) {
   return numCtx >= 1024 ? `${Math.floor(numCtx / 1024)}K` : String(numCtx);
 }
 function calculateExpandedVariantContextWindow(specs, modelSizeGB2, kvBytesPerToken, archMax) {
+  const memoryBudget = calculateContextWindow(
+    specs,
+    modelSizeGB2,
+    kvBytesPerToken,
+    archMax
+  );
   if (archMax && archMax > 0) {
-    const numCtx = Math.max(2048, Math.floor(archMax / 1024) * 1024);
+    const archCtx = Math.max(2048, Math.floor(archMax / 1024) * 1024);
+    const numCtx = Math.min(archCtx, memoryBudget.numCtx);
     return { numCtx, label: formatContextLabel(numCtx) };
   }
-  return calculateContextWindow(specs, modelSizeGB2, kvBytesPerToken, archMax);
+  return memoryBudget;
 }
 function ask(rl, question) {
   return new Promise((resolve52) => {
@@ -579437,7 +579498,10 @@ function getVenvDir2() {
 }
 function hasVenvModule() {
   try {
-    execSync50("python3 -m venv --help", { stdio: "pipe", timeout: 5e3 });
+    execSync50('python3 -c "import ensurepip, venv"', {
+      stdio: "pipe",
+      timeout: 5e3
+    });
     return true;
   } catch {
     return false;
@@ -579470,15 +579534,15 @@ function ensureVenv(log22) {
       }
     }
   }
-  log22("Creating Python venv for vision deps...");
   if (!hasCmd(pythonCmd) && !hasCmd("python3")) {
     log22(`${pythonCmd} not found — cannot create venv.`);
     return null;
   }
   if (!isWin2 && !hasVenvModule()) {
-    log22("python3 venv module not available — venv creation skipped.");
+    log22("python3-venv is not installed — venv creation skipped. Install with: sudo apt-get install python3-venv");
     return null;
   }
+  log22("Creating Python venv for vision deps...");
   try {
     mkdirSync52(join107(homedir34(), ".omnius"), { recursive: true });
     const pyCmd = hasCmd(pythonCmd) ? pythonCmd : "python3";
@@ -580034,7 +580098,10 @@ async function repairExpandedVariantIfStale(variantModel, fallbackBaseModel, bac
     return { repaired: false, currentNumCtx: state.currentNumCtx, baseModel: null, resolvedModel: variantName };
   }
   const canonicalModel = expandedModelName(baseModel);
-  const needsCtxRepair = !(state.currentNumCtx > 0 && state.currentNumCtx >= targetNumCtx);
+  const hasCtx = state.currentNumCtx > 0;
+  const tooSmall = hasCtx && state.currentNumCtx < targetNumCtx;
+  const tooLarge = hasCtx && state.currentNumCtx > Math.floor(targetNumCtx * 1.1);
+  const needsCtxRepair = !hasCtx || tooSmall || tooLarge;
   const needsCanonicalVariant = variantName !== canonicalModel;
   let repaired = false;
   if (needsCtxRepair) {
@@ -652800,15 +652867,41 @@ ${result.summary}`
     },
     () => new Promise((resolve52) => {
       depSudoPromptPending = true;
+      if (process.stdout.isTTY) {
+        process.stdout.write("\x1B[?1000l\x1B[?1002l\x1B[?1006l");
+      }
+      const origProvider = statusBar.inputStateProvider;
+      if (statusBar?.isActive) {
+        statusBar.inputStateProvider = () => {
+          const state = origProvider?.() ?? { line: "", cursor: 0 };
+          return { line: "●".repeat(state.line.length), cursor: state.cursor };
+        };
+      }
+      const footerPwPrompt = `\x1B[38;5;198m● password:\x1B[0m `;
+      const prevPromptText = statusBar.promptText;
+      const prevPromptWidth = statusBar.promptWidth;
+      if (statusBar?.isActive) {
+        statusBar.setPromptText(footerPwPrompt, 12);
+      }
       depSudoResolver = (pw2) => {
         depSudoPromptPending = false;
         depSudoResolver = null;
+        if (statusBar?.isActive) {
+          statusBar.inputStateProvider = origProvider ?? (() => ({ line: rl.line ?? "", cursor: rl.cursor ?? 0 }));
+          if (prevPromptText !== void 0 && prevPromptWidth !== void 0) {
+            statusBar.setPromptText(prevPromptText, prevPromptWidth);
+          }
+        }
+        if (process.stdout.isTTY) {
+          process.stdout.write("\x1B[?1002h\x1B[?1006h");
+        }
         if (pw2) sessionSudoPassword = pw2;
         resolve52(pw2);
       };
       const pwPrompt = `
-  ${c3.bold(c3.yellow("🔑 Password needed for dependency install:"))}
-  ${c3.dim("Type your sudo password below and press Enter.")}
+  \x1B[5;38;5;198m⚠ Password needed for dependency install\x1B[0m
+  ${c3.dim("Type your sudo password below and press Enter. Input is hidden.")}
+
 `;
       if (isNeovimActive()) {
         writeToNeovimOutput(pwPrompt);

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.104",
+  "version": "1.0.106",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.104",
+      "version": "1.0.106",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.104",
+  "version": "1.0.106",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",