@@ -70,9 +70,21 @@ jobs:
           personal_token: ${{ secrets.MODELSCRIPT_SITE_DEPLOY_PAT }}
           external_repository: modelscript/modelscript.github.io
           publish_dir: ./packages/site/build/client
+      - name: Cache WebLLM Model
+        id: cache-model
+        uses: actions/cache@v4
+        with:
+          path: packages/ide/models
+          key: ${{ runner.os }}-qwen3-0.6b-q4f16_1-mlc-v1
+      - name: Download WebLLM Model
+        if: steps.cache-model.outputs.cache-hit != 'true'
+        run: bash packages/ide/scripts/download-model.sh
       - name: Build IDE Static
         if: github.ref == 'refs/heads/main' && github.event_name != 'pull_request'
         run: npm run build-static --workspace=@modelscript/ide
+      - name: Copy model to IDE static output
+        if: github.ref == 'refs/heads/main' && github.event_name != 'pull_request'
+        run: cp -r packages/ide/models packages/ide/dist/static/api/models
       - name: Deploy IDE
         uses: peaceiris/actions-gh-pages@v4
         if: github.ref == 'refs/heads/main' && github.event_name != 'pull_request'
 
@@ -119,6 +119,14 @@ EXPOSE 80
 # ==============================================================================
 # IDE
 # ==============================================================================
+
+# Download WebLLM model weights (cached layer)
+FROM node:22-alpine AS download-model
+RUN apk add --no-cache curl bash
+WORKDIR /models
+COPY packages/ide/scripts/download-model.sh /tmp/download-model.sh
+RUN bash /tmp/download-model.sh /models
+
 FROM deps AS build-ide-false
 COPY scripts scripts
 COPY packages/core packages/core
@@ -164,6 +172,7 @@ COPY --from=build-ide /app/packages/vscode/syntaxes packages/vscode/syntaxes
 COPY --from=build-ide /app/packages/vscode/language-configuration.json packages/vscode/
 COPY --from=build-ide /app/packages/morsel/public packages/morsel/public
 COPY --from=build-ide /app/node_modules/@vscode node_modules/@vscode
+COPY --from=download-model /models packages/ide/models
 EXPOSE 3200
 ENV NODE_ENV=production
 ENV PORT=3200
 
@@ -92,6 +92,16 @@ This launches:
 | API                   | http://localhost:3000 | REST API for simulation, publishing, queries |
 | IDE (VS Code Web)     | http://localhost:3200 | Browser-based VS Code with Modelica support  |
 
+#### AI Chat (Optional)
+
+The IDE includes a browser-local AI assistant powered by WebLLM (Qwen3-0.6B). To enable it, download the model weights (~350 MB, one-time):
+
+```bash
+npm run download-model --workspace=@modelscript/ide
+```
+
+Once downloaded, restart `npm run dev` and open the **ModelScript AI** panel in the IDE sidebar.
+
 ### CLI Usage
 
 After building, the CLI is available as `msc`:
 
@@ -10,6 +10,7 @@
     "build-server": "npx esbuild src/server.ts --bundle --platform=node --format=esm --outfile=dist/server.js --external:express --external:http-proxy-middleware",
     "build-static": "npm run download-vscode && npm run build-github-fs && npx tsx src/build-static.ts",
     "dev": "npm run download-vscode && npx tsx src/server.ts",
+    "download-model": "bash scripts/download-model.sh",
     "download-vscode": "npx tsx src/download-vscode.ts",
     "lint": "eslint src --ext ts"
   },
 
@@ -0,0 +1,61 @@
+#!/usr/bin/env bash
+# Download Qwen3-0.6B-q4f16_1-MLC model weights and WASM for WebLLM.
+# Files are cached — re-running is a no-op if all files already exist.
+#
+# Usage: ./scripts/download-model.sh [target_dir]
+# Default target_dir: packages/ide/models
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+TARGET="${1:-$REPO_ROOT/models}"
+
+MODEL_DIR="$TARGET/Qwen3-0.6B-q4f16_1-MLC"
+WASM_FILE="$TARGET/Qwen3-0.6B-q4f16_1-ctx4k_cs1k-webgpu.wasm"
+
+HF_BASE="https://huggingface.co/mlc-ai/Qwen3-0.6B-q4f16_1-MLC/resolve/main"
+WASM_URL="https://huggingface.co/mlc-ai/binary-mlc-llm-libs/resolve/main/Qwen3-0.6B-q4f16_1-ctx4k_cs1k-webgpu.wasm"
+
+# Files to download from the HuggingFace model repo
+MODEL_FILES=(
+  mlc-chat-config.json
+  ndarray-cache.json
+  tokenizer.json
+  tokenizer_config.json
+  vocab.json
+  merges.txt
+  params_shard_0.bin
+  params_shard_1.bin
+  params_shard_2.bin
+  params_shard_3.bin
+  params_shard_4.bin
+  params_shard_5.bin
+  params_shard_6.bin
+  params_shard_7.bin
+  params_shard_8.bin
+)
+
+echo "==> Downloading Qwen3-0.6B-q4f16_1-MLC model to $TARGET"
+
+mkdir -p "$MODEL_DIR"
+
+# Download model files
+for file in "${MODEL_FILES[@]}"; do
+  if [ -f "$MODEL_DIR/$file" ]; then
+    echo "  [cached] $file"
+  else
+    echo "  [downloading] $file"
+    curl -fSL --retry 3 "$HF_BASE/$file" -o "$MODEL_DIR/$file"
+  fi
+done
+
+# Download WASM
+if [ -f "$WASM_FILE" ]; then
+  echo "  [cached] $(basename "$WASM_FILE")"
+else
+  echo "  [downloading] $(basename "$WASM_FILE")"
+  curl -fSL --retry 3 "$WASM_URL" -o "$WASM_FILE"
+fi
+
+echo "==> Model download complete"