omnius 1.0.432 → 1.0.433

package/dist/index.js

@@ -8216,6 +8216,7 @@ function runProcessBuffer(file, args = [], options2 = {}) {
       stdio: ["pipe", "pipe", "pipe"],
       windowsHide: true
     });
+    options2.onSpawn?.(child);
     const stdoutChunks = [];
     const stderrChunks = [];
     const maxBuffer = options2.maxBuffer ?? 16 * 1024 * 1024;
@@ -8223,6 +8224,27 @@ function runProcessBuffer(file, args = [], options2 = {}) {
     let stderrBytes = 0;
     let settled = false;
     let timedOut = false;
+    let aborted = false;
+    let killTimer = null;
+    const abortProcess = () => {
+      aborted = true;
+      try {
+        child.kill("SIGTERM");
+      } catch {
+      }
+      killTimer = setTimeout(() => {
+        try {
+          if (child.exitCode === null && child.signalCode === null)
+            child.kill("SIGKILL");
+        } catch {
+        }
+      }, 1500);
+      killTimer.unref?.();
+    };
+    if (options2.signal?.aborted)
+      abortProcess();
+    else
+      options2.signal?.addEventListener("abort", abortProcess, { once: true });
     const timer = options2.timeout ? setTimeout(() => {
       timedOut = true;
       try {
@@ -8237,6 +8259,9 @@ function runProcessBuffer(file, args = [], options2 = {}) {
       settled = true;
       if (timer)
         clearTimeout(timer);
+      if (killTimer)
+        clearTimeout(killTimer);
+      options2.signal?.removeEventListener("abort", abortProcess);
       fn(value2);
     };
     child.stdout?.on("data", (chunk) => {
@@ -8273,6 +8298,10 @@ function runProcessBuffer(file, args = [], options2 = {}) {
         finish(reject, new AsyncProcessError(`Process timed out after ${options2.timeout}ms: ${file}`, result));
         return;
       }
+      if (aborted) {
+        finish(reject, new AsyncProcessError(`Process aborted: ${file}`, result));
+        return;
+      }
       if (stdoutBytes > maxBuffer || stderrBytes > maxBuffer) {
         finish(reject, new AsyncProcessError(`Process exceeded maxBuffer: ${file}`, result));
         return;
@@ -555784,6 +555813,9 @@ var init_visual_trigger = __esm({
 import { existsSync as existsSync78, mkdirSync as mkdirSync47, readFileSync as readFileSync57, rmSync as rmSync10, writeFileSync as writeFileSync39 } from "node:fs";
 import { homedir as homedir28, tmpdir as tmpdir20 } from "node:os";
 import { basename as basename19, isAbsolute as isAbsolute8, join as join92, resolve as resolve46 } from "node:path";
+function isAbortSignal(value2) {
+  return Boolean(value2 && typeof value2 === "object" && "aborted" in value2 && typeof value2.addEventListener === "function");
+}
 function buildYolo26BootstrapPlan(model = DEFAULT_YOLO26_MODEL) {
   return {
     runtimeRoot: LIVE_MEDIA_ROOT,
@@ -556707,7 +556739,8 @@ var init_live_media_loop = __esm({
             const raw = await execFileText(runtime.python, [scriptPath2, cfgPath], {
               timeout: Math.max(12e4, Number(args["duration_sec"] ?? 8) * 1e3 + 9e5),
               maxBuffer: 64 * 1024 * 1024,
-              env: { ...process.env, PYTHONUNBUFFERED: "1" }
+              env: { ...process.env, PYTHONUNBUFFERED: "1" },
+              signal: isAbortSignal(args["abort_signal"]) ? args["abort_signal"] : void 0
             });
             const jsonLine2 = raw.trim().split("\n").reverse().find((line) => line.trim().startsWith("{") && line.trim().endsWith("}"));
             if (!jsonLine2)
@@ -604534,11 +604567,12 @@ var init_camera_streamer = __esm({
           }
           const killTimer = setTimeout(() => {
             try {
-              if (!child.killed) child.kill("SIGKILL");
+              if (child.exitCode === null && child.signalCode === null) child.kill("SIGKILL");
             } catch {
             }
           }, 1500);
           killTimer.unref?.();
+          child.once("exit", () => clearTimeout(killTimer));
           stream.process = null;
         }
         try {
@@ -604698,7 +604732,7 @@ var init_live_vlm = __esm({
                 model: this.model,
                 stream: false,
                 format: "json",
-                keep_alive: String(process.env["OMNIUS_LIVE_VLM_KEEP_ALIVE"] ?? "30s"),
+                keep_alive: String(process.env["OMNIUS_LIVE_VLM_KEEP_ALIVE"] ?? "5s"),
                 options: { temperature: 0.1, num_predict: 260 },
                 messages: [
                   { role: "system", content: VLM_SYSTEM_PROMPT },
@@ -606851,6 +606885,7 @@ var init_live_sensors = __esm({
       videoGeneration = 0;
       liveGpuYieldUntil = 0;
       resourceArbiter = getLiveResourceArbiter();
+      liveMediaAbortControllers = /* @__PURE__ */ new Set();
       videoInFlight = false;
       audioInFlight = false;
       audioActivityInFlight = false;
@@ -607187,7 +607222,7 @@ var init_live_sensors = __esm({
         const now2 = Date.now();
         if (now2 >= this.liveGpuYieldUntil) {
           this.liveGpuYieldUntil = Math.max(decision2.until ?? 0, now2 + 15e3);
-          void this.liveVlm.unloadModel().catch(() => false);
+          void this.releaseLiveGpuMemory("live vision release", "vision").catch(() => false);
           this.emitFeedbackThrottled(`live-video-gpu-yield:${kind}`, {
             kind: "status",
             title: "Live video GPU yield",
@@ -607342,7 +607377,7 @@ var init_live_sensors = __esm({
           this.videoGeneration++;
           this.lastStreamFrameMtime.clear();
           this.cameraStreamers?.stopAll();
-          void this.liveVlm.unloadModel().catch(() => false);
+          void this.releaseLiveGpuMemory("live vision release", "vision").catch(() => false);
         }
         this.persist();
         this.clearLoopTimers();
@@ -607375,8 +607410,10 @@ var init_live_sensors = __esm({
         this.lastInferenceAt.clear();
         this.autoOrientationInFlight.clear();
         this.liveSpeechInFlight.clear();
+        for (const controller of this.liveMediaAbortControllers) controller.abort();
+        this.liveMediaAbortControllers.clear();
         this.resourceArbiter.releaseOwnerPrefix("live-");
-        void this.liveVlm.unloadModel().catch(() => false);
+        void this.releaseLiveGpuMemory("live stop", this.liveStopOllamaUnloadMode()).catch(() => false);
         this.snapshot = {
           version: 1,
           repoRoot: this.repoRoot,
@@ -607390,6 +607427,40 @@ var init_live_sensors = __esm({
         this.emitStatus();
         this.emitDashboard();
       }
+      liveStopOllamaUnloadMode() {
+        const raw = String(process.env["OMNIUS_LIVE_STOP_UNLOAD_OLLAMA"] ?? "").trim().toLowerCase();
+        if (raw === "0" || raw === "off" || raw === "false" || raw === "none") return "off";
+        if (raw === "vision" || raw === "vlm" || raw === "video") return "vision";
+        if (raw === "all" || raw === "1" || raw === "true" || raw === "yes") return "all";
+        return isJetsonUnifiedMemoryHost() ? "all" : "vision";
+      }
+      keptOllamaModels() {
+        const keep = String(process.env["OMNIUS_LIVE_KEEP_OLLAMA_MODELS"] ?? "").split(",").map((entry) => entry.trim()).filter(Boolean);
+        return new Set(keep);
+      }
+      shouldUnloadOllamaModelForLiveCleanup(modelName, mode, keep) {
+        if (!modelName) return false;
+        if (keep.has(modelName) || keep.has(modelName.split(":")[0] ?? modelName)) return false;
+        if (mode === "all") return true;
+        if (modelName === this.liveVlm.model) return true;
+        return /(?:moondream|minicpm|llava|bakllava|vision|clip|florence|qwen.*vl|gemma.*vision)/i.test(modelName);
+      }
+      async releaseLiveGpuMemory(reason, mode = "vision") {
+        await this.liveVlm.unloadModel().catch(() => false);
+        if (mode === "off") return;
+        const broker = getModelBroker();
+        const snapshot = await broker.pollOnce().catch(() => broker.snapshot());
+        const keep = this.keptOllamaModels();
+        const targets = /* @__PURE__ */ new Set();
+        for (const model of snapshot.loaded ?? []) {
+          if (model.host !== "ollama") continue;
+          if (this.shouldUnloadOllamaModelForLiveCleanup(model.name, mode, keep)) targets.add(model.name);
+        }
+        for (const modelName of targets) {
+          await broker.unloadOllamaModel(modelName, reason).catch(() => false);
+        }
+        globalThis.gc?.();
+      }
       startRunMode(options2 = {}) {
         this.agentActionEnabled = Boolean(options2.agent);
         this.liveSpeechEnabled = Boolean(options2.speech ?? options2.agent);
@@ -607677,8 +607748,11 @@ var init_live_sensors = __esm({
           this.lastInferenceAt.set(source, sampledAt);
           const loop = new LiveMediaLoopTool(this.repoRoot);
           const inferFromCapturedFrame = preview.ok && Boolean(preview.framePath);
+          const mediaAbortController = new AbortController();
+          this.liveMediaAbortControllers.add(mediaAbortController);
           const result = await loop.execute({
             action: "watch",
+            abort_signal: mediaAbortController.signal,
             source_kind: inferFromCapturedFrame ? "file" : "camera",
             ...inferFromCapturedFrame ? { path: preview.framePath } : { camera: source },
             yolo_family: "yoloe-26",
@@ -607697,6 +607771,8 @@ var init_live_sensors = __esm({
             max_segments_per_frame: 8,
             crop_faces: true,
             recognize_faces: true
+          }).finally(() => {
+            this.liveMediaAbortControllers.delete(mediaAbortController);
           });
           if (!this.isVideoWorkCurrent(generation)) return "Video context is disabled.";
           inference = result.success ? String(result.output || result.llmContent || "") : `Video inference failed: ${result.error || result.output || "unknown error"}`;

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.432",
+  "version": "1.0.433",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.432",
+      "version": "1.0.433",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.432",
+  "version": "1.0.433",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",