open-agents-ai 0.185.40 → 0.185.42

package/dist/index.js

@@ -41395,7 +41395,7 @@ function execAsync(cmd, opts = {}) {
 function selectWeightTier(vramGB) {
   if (vramGB >= 48)
     return "original";
-  if (vramGB >= 16)
+  if (vramGB >= 24)
     return "nf4-distilled";
   return "nf4";
 }
@@ -41499,13 +41499,20 @@ function isPersonaPlexInstalled() {
   return existsSync37(join54(PERSONAPLEX_DIR, "model_ready"));
 }
 function getWeightTier() {
+  const detected = detectPersonaPlexCapability();
   const tierFile = join54(PERSONAPLEX_DIR, "weight_tier");
   if (existsSync37(tierFile)) {
     const saved = readFileSync28(tierFile, "utf8").trim();
-    if (saved in WEIGHT_REPOS)
+    if (saved in WEIGHT_REPOS) {
+      const vram = detected.vramGB;
+      if (saved === "nf4-distilled" && vram < 24) {
+        writeFileSync16(tierFile, "nf4");
+        return "nf4";
+      }
       return saved;
+    }
   }
-  return detectPersonaPlexCapability().weightTier;
+  return detected.weightTier;
 }
 function getWeightRepoInfo(tier) {
   return WEIGHT_REPOS[tier];
@@ -41937,7 +41944,12 @@ print('Converted')
   } catch {
     log("Ollama not detected \u2014 running PersonaPlex standalone (no hybrid)");
   }
-  log(`Starting PersonaPlex daemon (${tier} tier${hybridEnabled ? ", hybrid" : ""})...`);
+  const caps = detectPersonaPlexCapability();
+  const needsOffload = caps.vramGB > 0 && caps.vramGB < 24;
+  if (needsOffload) {
+    log(`GPU has ${caps.vramGB.toFixed(0)}GB VRAM \u2014 enabling CPU offload (model needs ~19GB)`);
+  }
+  log(`Starting PersonaPlex daemon (${tier} tier${hybridEnabled ? ", hybrid" : ""}${needsOffload ? ", cpu-offload" : ""})...`);
   const serverArgs = [
     "-m",
     "moshi.server",
@@ -41953,6 +41965,8 @@ print('Converted')
   ];
   if (hybridEnabled)
     serverArgs.push("--hybrid");
+  if (needsOffload)
+    serverArgs.push("--cpu-offload");
   const serverEnv = { ...process.env };
   if (hybridEnabled) {
     serverEnv["HYBRID_ENABLED"] = "1";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.185.40",
+  "version": "0.185.42",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",