Caching slash-loaded models to use.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "localm",
-  "version": "1.1.19",
+  "version": "1.1.20",
   "description": "Chat application",
   "scripts": {
     "build": "esbuild src/index.js --target=es6 --bundle --sourcemap --outfile=./index.js --format=iife --external:fs --external:path --external:child_process --external:ws --external:katex/dist/katex.min.css",

src/app/model-list.js

@@ -92,15 +92,16 @@ export async function fetchBrowserModels() {
           hasTokenizer: !!hasTokenizer,
           missingFiles: !!missingFiles,
           missingReason: missingReason || '',
-          downloads: m.downloads || 0
+      downloads: m.downloads || 0,
+      tags: Array.isArray(m.tags) ? m.tags.slice() : []
         });
       } catch (e) {
         return null;
       }
     }).filter(m => m !== null);
 
-    // Keep only models that have both ONNX and tokenizer files
-    const withFiles = processed.filter(p => p && p.hasOnnx && p.hasTokenizer);
+  // Keep only models that have both ONNX and tokenizer files AND support chat
+  const withFiles = processed.filter(p => p && p.hasOnnx && p.hasTokenizer && isModelChatCapable(p));
 
   // Sort by downloads desc
   withFiles.sort((a, b) => ((b && b.downloads) || 0) - ((a && a.downloads) || 0));
@@ -343,6 +344,31 @@ function detectRequiredFiles(model) {
   return { hasOnnx, hasTokenizer, missingFiles: missing, missingReason: reason };
 }
 
+/**
+ * Determine if a model supports chat-style inputs/outputs.
+ * Uses pipeline_tag, tags, and name heuristics as fallback.
+ * @param {any} model
+ */
+function isModelChatCapable(model) {
+  if (!model) return false;
+  const allowedPipelines = new Set(['text-generation', 'conversational', 'text2text-generation', 'chat']);
+  if (model.pipeline_tag && allowedPipelines.has(model.pipeline_tag)) return true;
+  // tags array may contain 'conversational' or 'chat'
+  if (Array.isArray(model.tags)) {
+    for (const t of model.tags) {
+      if (typeof t === 'string' && allowedPipelines.has(t)) return true;
+    }
+  }
+  // fallback heuristics in id/name: look for chat, conversational, dialog, instruct
+  const id = (model.id || '').toLowerCase();
+  const name = (model.name || '').toLowerCase();
+  const heuristics = ['chat', 'conversational', 'dialog', 'instruct', 'instruction'];
+  for (const h of heuristics) {
+    if (id.includes(h) || name.includes(h)) return true;
+  }
+  return false;
+}
+
 /**
  * Get fallback models if API fetch fails
  * @returns {ModelInfo[]}

src/worker/boot-worker.js

@@ -4,6 +4,7 @@ import { ModelCache } from './model-cache';
 
 export function bootWorker() {
   const modelCache = new ModelCache();
+  let selectedModel = modelCache.knownModels[0];
   // Report starting
   try {
     self.postMessage({ type: 'status', status: 'initializing' });
@@ -29,6 +30,7 @@ export function bootWorker() {
         const { modelName = modelCache.knownModels[0] } = data;
         try {
           const pipe = await modelCache.getModel({ modelName });
+          selectedModel = modelName;
           self.postMessage({ id, type: 'response', result: { model: modelName, status: 'loaded' } });
         } catch (err) {
           self.postMessage({ id, type: 'error', error: String(err) });
@@ -43,7 +45,7 @@ export function bootWorker() {
     }
   }
 
-  async function handleRunPrompt({ prompt, modelName = modelCache.knownModels[0], id, options }) {
+  async function handleRunPrompt({ prompt, modelName = selectedModel, id, options }) {
     try {
       const pipe = await modelCache.getModel({ modelName });
       // run the pipeline