getmy-ruflo 3.5.55 → 3.5.56

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "getmy-ruflo",
-  "version": "3.5.55",
+  "version": "3.5.56",
   "description": "GetMy Ruflo - AI agent orchestration platform with real terminal execution, 259 MCP tools, 60+ agents, and swarm coordination",
   "main": "bin/ruflo.js",
   "type": "module",

package/v3/@claude-flow/cli/dist/src/mcp-tools/ollama-tools.js

@@ -6,7 +6,7 @@
  */
 function getOllamaHost() {
     const host = (typeof process !== "undefined" && process.env["OLLAMA_HOST"]) ||
-        "http://localhost:11434";
+        "http://192.168.0.85:11434";
     return host.trim();
 }
 function ok(data) {
@@ -34,7 +34,7 @@ export const ollamaTools = [
     {
         name: "ollama_route",
         description: "Auto-route a task to the best available Ollama model based on complexity. " +
-            "Automatically selects qwen2.5-coder:7b for short tasks or qwen2.5-coder:32b for longer ones. " +
+            "Automatically selects qwen3-coder:latest for short tasks or qwen3-coder:latest for longer ones. " +
             "Returns the model response or an error if Ollama is unavailable (caller should fall back to Claude).",
         inputSchema: {
             type: "object",
@@ -68,23 +68,18 @@ export const ollamaTools = [
                 const models = (tags.models ?? []).map((m) => m.name);
                 if (models.length === 0)
                     return fail("No Ollama models loaded — fall back to Claude");
-                // Select model
-                const has7b = models.some((m) => m.includes("7b"));
-                const has32b = models.some((m) => m.includes("32b"));
+                // Select model — prefer qwen3-coder (primary Mac Mini model), then qwen2.5 variants
+                const qwen3 = models.find((m) => m.includes("qwen3"));
+                const qwen7b = models.find((m) => m.includes("qwen") && m.includes("7b"));
+                const qwen32b = models.find((m) => m.includes("qwen") && m.includes("32b"));
                 let model;
-                if (preferSmall && has7b) {
-                    model = models.find((m) => m.includes("7b"));
-                }
-                else if (prompt.length < 200 && has7b) {
-                    model = models.find((m) => m.includes("7b"));
-                }
-                else if (has32b) {
-                    model = models.find((m) => m.includes("32b"));
+                if (preferSmall && qwen7b) {
+                    model = qwen7b;
                 }
                 else {
-                    model = models[0];
+                    model = qwen3 || qwen32b || qwen7b || models[0];
                 }
-                // Query — use timeout for large models (qwen2.5-coder:32b can take 60-120s)
+                // Query — use timeout for large models (qwen3-coder:latest can take 60-120s)
                 const timeoutMs = input.timeout || 300000;
                 const response = await fetch(`${host}/api/generate`, {
                     method: "POST",
@@ -170,7 +165,7 @@ export const ollamaTools = [
                 },
                 model: {
                     type: "string",
-                    description: "Ollama model (default: qwen2.5-coder:32b)",
+                    description: "Ollama model (default: qwen3-coder:latest)",
                 },
                 apply: {
                     type: "boolean",
@@ -183,7 +178,7 @@ export const ollamaTools = [
         tags: ["ollama", "llm", "local-ai", "github", "code-generation"],
         handler: async (input) => {
             const issueNum = input.issue_number;
-            const model = input.model || "qwen2.5-coder:32b";
+            const model = input.model || "qwen3-coder:latest";
             const host = getOllamaHost();
             try {
                 // 1. Get issue details via gh CLI
@@ -235,8 +230,8 @@ export const ollamaTools = [
     {
         name: "ollama_pipeline",
         description: "Full zero-cost issue implementation pipeline using local Ollama. " +
-            "Lists open GitHub issues, uses qwen2.5-coder:7b to identify files, " +
-            "qwen2.5-coder:32b to generate code, then creates branches, commits, and PRs. " +
+            "Lists open GitHub issues, uses qwen3-coder:latest to identify files, " +
+            "qwen3-coder:latest to generate code, then creates branches, commits, and PRs. " +
             "Costs $0 — all inference runs locally on Ollama.",
         inputSchema: {
             type: "object",
@@ -251,11 +246,11 @@ export const ollamaTools = [
                 },
                 model: {
                     type: "string",
-                    description: "Ollama model for code generation (default: qwen2.5-coder:32b)",
+                    description: "Ollama model for code generation (default: qwen3-coder:latest)",
                 },
                 analysis_model: {
                     type: "string",
-                    description: "Ollama model for file analysis (default: qwen2.5-coder:7b)",
+                    description: "Ollama model for file analysis (default: qwen3-coder:latest)",
                 },
                 limit: {
                     type: "number",
@@ -284,8 +279,8 @@ export const ollamaTools = [
         tags: ["ollama", "llm", "local-ai", "github", "pipeline", "automation"],
         handler: async (input) => {
             const host = getOllamaHost();
-            const codeModel = input.model || "qwen2.5-coder:32b";
-            const analysisModel = input.analysis_model || "qwen2.5-coder:7b";
+            const codeModel = input.model || "qwen3-coder:latest";
+            const analysisModel = input.analysis_model || "qwen3-coder:latest";
             const baseBranch = input.base_branch || "main";
             const limit = input.limit || 10;
             const skipIssues = input.skip_issues || [];

package/v3/@claude-flow/cli/dist/src/ruvector/enhanced-model-router.js

@@ -255,21 +255,20 @@ export class EnhancedModelRouter {
      * Prefers 7b for short/simple prompts, 32b for longer ones.
      */
     selectOllamaModel(task) {
-        // Prefer qwen3-coder (MoE, best quality), then qwen2.5-coder variants
-        const qwen3 = this.ollamaModels.find((m) => m.includes('qwen3'));
+        // Prefer Q4_K_M quants (fast load, ~66 tok/s on M4 Pro) over Q8_0 (too slow to load)
+        const qwen3q4 = this.ollamaModels.find((m) => m.includes('qwen3') && (m.includes('q4') || m === 'qwen3-coder:latest'));
+        const qwen3 = this.ollamaModels.find((m) => m.includes('qwen3') && !m.includes('q8'));
         const qwen7b = this.ollamaModels.find((m) => m.includes('qwen') && m.includes('7b'));
-        const qwen32b = this.ollamaModels.find((m) => m.includes('qwen') && m.includes('32b'));
-        // qwen3-coder is the primary model on Mac Mini — always prefer it
+        // qwen3-coder Q4_K_M is the preferred model — fast load, good quality
+        if (qwen3q4)
+            return qwen3q4;
         if (qwen3)
             return qwen3;
-        // Fallback to qwen2.5-coder size variants
         if (task.length < 200 && qwen7b)
             return qwen7b;
-        if (qwen32b)
-            return qwen32b;
         if (qwen7b)
             return qwen7b;
-        return (this.ollamaModels[0] ?? 'qwen3-coder:30b-ctx32k');
+        return (this.ollamaModels[0] ?? 'qwen3-coder:latest');
     }
     /**
      * Check if a task is suitable for Ollama.