omnius 1.0.520 → 1.0.521

package/dist/index.js

@@ -566866,6 +566866,9 @@ var init_ollama_pool = __esm({
           return this.buildSlot(pick, resolvedAgentId);
         }
         if (placementMode === "constrained") {
+          if (optsWithAgent.queuePolicy === "skip") {
+            throw new Error("No GPU slot immediately available. All slots occupied.");
+          }
           return this.acquireQueued(optsWithAgent, resolvedAgentId);
         }
         const spawned = placementMode === "elastic" ? await this.maybeSpawnInstance(optsWithAgent.model) : null;
@@ -566875,6 +566878,9 @@ var init_ollama_pool = __esm({
             this.recordAffinity(resolvedAgentId, spawned.state.id);
           return this.buildSlot(spawned, resolvedAgentId);
         }
+        if (optsWithAgent.queuePolicy === "skip") {
+          throw new Error("No GPU slot immediately available. All slots occupied.");
+        }
         return this.acquireQueued(optsWithAgent, resolvedAgentId);
       }
       /** Synchronous routing decision; returns the instance or null if every one is saturated. */
@@ -566936,7 +566942,8 @@ var init_ollama_pool = __esm({
             return this.buildSlot(pick, agentId);
           }
           await new Promise((resolve79, reject) => {
-            const timer = setTimeout(() => reject(new Error("No GPU slot available after 30s. All slots occupied.")), 3e4);
+            const timeoutMs = Number.isFinite(opts.queueTimeoutMs) && (opts.queueTimeoutMs ?? 0) > 0 ? Math.floor(opts.queueTimeoutMs) : 3e4;
+            const timer = setTimeout(() => reject(new Error(`No GPU slot available after ${(timeoutMs / 1e3).toFixed(0)}s. All slots occupied.`)), timeoutMs);
             this.slotWaiters.push(() => {
               clearTimeout(timer);
               resolve79();
@@ -603790,7 +603797,9 @@ ${description}`
           body["num_ctx"] = reqNumCtx;
         }
         let poolSlot = shouldUseOllamaPoolForBaseUrl(this.baseUrl) ? await getOllamaPool({ baseInstanceUrl: this.baseUrl }).acquire({
-          model: this.model
+          model: this.model,
+          queuePolicy: request.poolQueuePolicy,
+          queueTimeoutMs: request.poolQueueTimeoutMs
         }) : null;
         let requestBaseUrl = poolSlot?.baseUrl ?? this.baseUrl;
         let poolSuccess = false;
@@ -604185,7 +604194,9 @@ ${description}`
           body["num_ctx"] = reqNumCtx;
         }
         let poolSlot = shouldUseOllamaPoolForBaseUrl(this.baseUrl) ? await getOllamaPool({ baseInstanceUrl: this.baseUrl }).acquire({
-          model: this.model
+          model: this.model,
+          queuePolicy: request.poolQueuePolicy,
+          queueTimeoutMs: request.poolQueueTimeoutMs
         }) : null;
         let requestBaseUrl = poolSlot?.baseUrl ?? this.baseUrl;
         let poolSuccess = false;
@@ -686381,13 +686392,70 @@ var init_emotion_engine = __esm({
         this.state.arousal = BASELINE_AROUSAL + (this.state.arousal - BASELINE_AROUSAL) * decayFactor;
         this.state.updatedAt = Date.now();
       }
+      emotionModelInferencePolicy() {
+        const raw = process.env["OMNIUS_EMOTION_MODEL_INFERENCE"] ?? process.env["OMNIUS_EMOTION_LLM"] ?? this.config.modelInference ?? "auto";
+        const value2 = String(raw).trim().toLowerCase();
+        if (["0", "false", "off", "no", "disabled"].includes(value2)) {
+          return "off";
+        }
+        if (["1", "true", "on", "yes", "enabled"].includes(value2)) {
+          return "on";
+        }
+        return "auto";
+      }
+      async admitInternalInference() {
+        const policy = this.emotionModelInferencePolicy();
+        if (policy === "off") {
+          return { allowed: false, reason: "emotion_model_inference_disabled" };
+        }
+        if (policy === "on") return { allowed: true };
+        if (!shouldUseOllamaPoolForBaseUrl(this.config.backendUrl)) {
+          return { allowed: true };
+        }
+        try {
+          const status = await getOllamaPool({
+            baseInstanceUrl: this.config.backendUrl
+          }).status();
+          const gpuCount = status.hardware.gpus.length;
+          if (status.placement.mode === "constrained") {
+            return {
+              allowed: false,
+              reason: `local_ollama_constrained_placement:${gpuCount}gpu`
+            };
+          }
+          const hasImmediateSlot = status.instances.some((inst) => {
+            const maxParallel = Math.max(1, inst.maxParallel);
+            return inst.inflight < maxParallel;
+          });
+          if (!hasImmediateSlot) {
+            return {
+              allowed: false,
+              reason: "local_ollama_no_immediate_background_slot"
+            };
+          }
+        } catch (err) {
+          const detail = err instanceof Error ? err.message : String(err);
+          return {
+            allowed: false,
+            reason: `local_ollama_pool_status_unavailable:${detail.slice(0, 160)}`
+          };
+        }
+        return { allowed: true };
+      }
       async runDirectInternalInference(opts) {
-        const backend = new OllamaAgenticBackend(
-          this.config.backendUrl,
-          this.config.model,
-          this.config.apiKey,
-          false
-        );
+        const cwd4 = resolve68(process.cwd());
+        const startedAt2 = Date.now();
+        const admission = await this.admitInternalInference();
+        if (!admission.allowed) {
+          this.recordInternalDecision({
+            stage: opts.stage,
+            status: "skipped",
+            elapsedMs: Date.now() - startedAt2,
+            skipReason: admission.reason,
+            promptChars: opts.system.length + opts.user.length
+          });
+          return null;
+        }
         const request = {
           messages: [
             { role: "system", content: opts.system },
@@ -686398,10 +686466,10 @@ var init_emotion_engine = __esm({
           maxTokens: opts.maxTokens,
           timeoutMs: opts.timeoutMs,
           think: false,
-          disableEmptyContentRecovery: true
+          disableEmptyContentRecovery: true,
+          poolQueuePolicy: "skip",
+          poolQueueTimeoutMs: 1
         };
-        const cwd4 = resolve68(process.cwd());
-        const startedAt2 = Date.now();
         recordContextWindowDump({
           source: "emotionEngine",
           stage: opts.stage,
@@ -686416,6 +686484,12 @@ var init_emotion_engine = __esm({
           request
         });
         try {
+          const backend = new OllamaAgenticBackend(
+            this.config.backendUrl,
+            this.config.model,
+            this.config.apiKey,
+            false
+          );
           const result = await backend.chatCompletion(request);
           const output = String(
             result.choices?.[0]?.message?.content ?? ""
@@ -686478,6 +686552,12 @@ var init_emotion_engine = __esm({
             timeoutMs: 1e4,
             note: "best-effort emotion label refresh"
           });
+          if (output === null) {
+            this.lastLabelUpdate = Date.now();
+            this.lastLabelValence = this.state.valence;
+            this.lastLabelArousal = this.state.arousal;
+            return;
+          }
           const match = output.match(new RegExp("^(\\p{Emoji_Presentation}|\\p{Emoji}\\uFE0F?)\\s+(\\S+)", "u"));
           if (match) {
             this.state.emoji = match[1];
@@ -686552,6 +686632,7 @@ var init_emotion_engine = __esm({
             timeoutMs: 2e4,
             note: "best-effort Telegram admin outreach gate"
           });
+          if (output === null) return;
           const message2 = this.normalizeReflectionDecision(output);
           if (message2) outreach(message2);
         } catch {

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.520",
+  "version": "1.0.521",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.520",
+      "version": "1.0.521",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.520",
+  "version": "1.0.521",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",