Improve CLI interaction flow

.dockerignore

@@ -0,0 +1,9 @@
+workspace/
+models/
+node_modules/
+*.egg-info/
+__pycache__/
+.git/
+*.pyc
+.env
+dist/

.github/workflows/build-windows.yml

@@ -0,0 +1,33 @@
+name: Build Windows Desktop
+
+on:
+  workflow_dispatch:  # manual trigger from GitHub UI
+  push:
+    paths:
+      - 'desktop/**'
+      - '.github/workflows/build-windows.yml'
+
+jobs:
+  build:
+    runs-on: windows-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+
+      - name: Install dependencies
+        run: |
+          pip install pyinstaller pywebview websockets httpx pyyaml ddgs pillow
+
+      - name: Build .exe
+        working-directory: desktop
+        run: |
+          pyinstaller launcher.py --name=Tsunami --onefile --windowed --add-data "index.html;." --add-data "ws_bridge.py;." --add-data "../tsunami;tsunami" --add-data "../config.yaml;." --add-data "../scaffolds;scaffolds" --hidden-import=websockets --hidden-import=webview --hidden-import=tsunami --hidden-import=tsunami.agent --hidden-import=tsunami.config --hidden-import=tsunami.tools
+
+      - name: Upload artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: Tsunami-Windows
+          path: desktop/dist/Tsunami.exe

.gitignore

@@ -39,3 +39,4 @@ Thumbs.db
 intel/
 workspace/
 *.output
+.vite/

Dockerfile

@@ -0,0 +1,61 @@
+# TSUNAMI — Containerized Autonomous Agent
+# Works on Mac (Apple Silicon + Intel), Linux, Windows (WSL2)
+#
+# Build:  docker build -t tsunami .
+# Run:    docker run -p 9876:9876 -p 8090:8090 tsunami "build me a calculator"
+# UI:     open http://localhost:9876 after it builds something
+#
+# With GPU (NVIDIA):
+#   docker run --gpus all -p 9876:9876 -p 8090:8090 tsunami "build me a game"
+#
+# Persist builds across runs:
+#   docker run -v tsunami-workspace:/app/workspace -p 9876:9876 tsunami "build tetris"
+
+FROM node:22-slim AS base
+
+# System deps
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    python3 python3-pip python3-venv \
+    git curl cmake build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+# --- Build llama.cpp from source (CPU by default, CUDA if available) ---
+RUN git clone --depth 1 https://github.com/ggerganov/llama.cpp /tmp/llama.cpp \
+    && cd /tmp/llama.cpp \
+    && cmake -B build -DGGML_NATIVE=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF \
+    && cmake --build build --config Release -j$(nproc) --target llama-server \
+    && cp build/bin/llama-server /usr/local/bin/llama-server \
+    && rm -rf /tmp/llama.cpp
+
+# --- Python deps ---
+COPY requirements.txt* ./
+RUN python3 -m pip install --break-system-packages --no-cache-dir \
+    httpx pyyaml ddgs pillow primp websockets \
+    && python3 -m pip install --break-system-packages --no-cache-dir playwright \
+    && python3 -m playwright install --with-deps chromium 2>/dev/null || true
+
+# --- Copy source ---
+COPY . /app/
+
+# --- Download models at build time (cached in image layer) ---
+RUN mkdir -p /app/models \
+    && echo "Downloading 2B eddy model..." \
+    && curl -fSL -o /app/models/Qwen3.5-2B-Q4_K_M.gguf \
+       "https://huggingface.co/unsloth/Qwen3.5-2B-GGUF/resolve/main/Qwen3.5-2B-Q4_K_M.gguf" \
+    && echo "Downloading 9B wave model..." \
+    && curl -fSL -o /app/models/Qwen3.5-9B-Q4_K_M.gguf \
+       "https://huggingface.co/unsloth/Qwen3.5-9B-GGUF/resolve/main/Qwen3.5-9B-Q4_K_M.gguf"
+
+# --- Workspace volume ---
+RUN mkdir -p /app/workspace/deliverables
+
+# Ports: wave(8090) eddy(8092) serve(9876) ws-bridge(3002)
+EXPOSE 8090 8092 9876 3002
+
+# --- Entrypoint: start everything and run the task ---
+COPY docker-entrypoint.sh /docker-entrypoint.sh
+RUN chmod +x /docker-entrypoint.sh
+
+ENTRYPOINT ["/docker-entrypoint.sh"]