Enhance remote deployment and update LLM functionality with tests

.dockerignore

@@ -0,0 +1,15 @@
+.git
+.github
+.pytest_cache
+.venv
+.tmp
+.codex
+__pycache__
+*.pyc
+*.pyo
+*.pyd
+data/*.db
+tests
+assets
+models
+app_config.toml

DEPLOYMENT.md

@@ -0,0 +1,63 @@
+# Deployment
+
+## Production baseline
+
+1. Copy `app_config_example.toml` to `app_config.toml`.
+2. Set `THINKING_GRAPH_SECRET_KEY` to a strong random value.
+3. Keep `debug = false`, `enable_cors = false`, `allow_runtime_settings_write = false`, and `allow_db_fallback = false`.
+4. Mount `./data` as a persistent volume.
+
+## User isolation
+
+The application now stores all nodes, connections, audits, and saved graphs with an `owner_id`.
+
+- Default mode: session isolation. Each browser session gets its own graph space.
+- Recommended production mode: proxy identity isolation. Configure your reverse proxy or auth gateway to inject a trusted user header, then set `trusted_identity_header` in `app_config.toml` or `THINKING_GRAPH_TRUSTED_IDENTITY_HEADER`.
+
+Example:
+
+```toml
+[auth]
+trusted_identity_header = "X-Forwarded-User"
+session_cookie_secure = true
+```
+
+Important:
+
+- Your reverse proxy must strip any client-supplied `X-Forwarded-User` header and set it itself.
+- Runtime settings writes are disabled by default so normal users cannot change shared server-side LLM config.
+- `/api/settings` no longer returns API keys in plaintext.
+
+## Docker
+
+Build:
+
+```bash
+docker build -t thinking-graph:latest .
+```
+
+Run:
+
+```bash
+docker run -d \
+  --name thinking-graph \
+  -p 5000:5000 \
+  -e THINKING_GRAPH_SECRET_KEY='replace-with-a-long-random-secret' \
+  -e THINKING_GRAPH_SESSION_COOKIE_SECURE=true \
+  -v $(pwd)/data:/app/data \
+  -v $(pwd)/app_config.toml:/app/app_config.toml:ro \
+  thinking-graph:latest
+```
+
+If you terminate TLS at a reverse proxy, also set:
+
+```bash
+-e APP_TRUSTED_PROXY_HOPS=1
+```
+
+## Reverse proxy checklist
+
+- Terminate HTTPS before exposing the app publicly.
+- Forward `X-Forwarded-Proto`, `X-Forwarded-Host`, and optionally `X-Forwarded-User`.
+- If using header-based identity, strip any incoming `X-Forwarded-User` from the public internet first.
+- Keep the app behind the proxy; do not expose Flask debug mode.

Dockerfile

@@ -0,0 +1,23 @@
+FROM python:3.11-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    APP_HOST=0.0.0.0 \
+    APP_PORT=5000
+
+WORKDIR /app
+
+RUN addgroup --system app && adduser --system --ingroup app app
+
+COPY requirements.txt ./
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+RUN mkdir -p /app/data && chown -R app:app /app
+
+USER app
+
+EXPOSE 5000
+
+CMD ["sh", "-c", "gunicorn --workers=2 --threads=4 --worker-class=gthread --bind=0.0.0.0:${APP_PORT:-5000} wsgi:app"]

QUICK_START_REFACTOR.md

@@ -0,0 +1,142 @@
+# Quick Start - Post-Refactoring
+
+## What Changed?
+
+The LLM integration has been completely refactored for better maintainability, testability, and extensibility. **All existing APIs work exactly the same** - no breaking changes!
+
+## Installation (Choose One)
+
+### Option 1: Cloud LLM (Recommended for Most Users)
+```bash
+pip install -r requirements.txt
+```
+Includes Flask + OpenAI API client. Use with DeepSeek, OpenAI, Claude, etc.
+
+### Option 2: Local LLM with NPU
+```bash
+pip install -r requirements-local-llm.txt
+```
+Includes ONNX Runtime / OpenVINO for local inference.
+
+### Option 3: Development Environment
+```bash
+pip install -r requirements-dev.txt
+```
+Adds pytest, httpx for testing.
+
+### Option 4: Everything
+```bash
+pip install -r requirements/all.txt
+```
+
+## Running Tests
+
+```bash
+python -m pytest tests/test_llm_refactor.py -v
+```
+
+This runs 30+ tests covering:
+- Schema validation
+- Generation pipeline
+- Review pipeline
+- Backend adapters
+
+## Using the New Architecture (Optional)
+
+### Old Way (Still Works!)
+```python
+from backend.services.llm_service import LLMService
+
+service = LLMService()
+result = service.generate_graph_from_topic("AI ethics")
+# Returns dict with nodes, connections, etc.
+```
+
+### New Way (Recommended for New Code)
+```python
+from config import LLMConfig
+from backend.services.llm_backends import create_llm_backend
+from backend.services.llm_graph_generation import GraphGenerationPipeline
+
+# Create backend
+config = LLMConfig.from_env()
+backend = create_llm_backend(
+    config=config,
+    backend_type="remote_api",
+    api_key="your-api-key",
+)
+
+# Create pipeline
+pipeline = GraphGenerationPipeline(backend)
+
+# Generate graph
+result = pipeline.generate(
+    topic="AI ethics",
+    max_nodes=15,
+    language="en"
+)
+
+# Access structured data
+if result.enabled and result.draft:
+    for node in result.draft.nodes:
+        print(f"{node.id}: {node.content}")
+
+    for conn in result.draft.connections:
+        print(f"{conn.source_id} -> {conn.target_id} ({conn.conn_type})")
+```
+
+### Benefits of New Approach
+- ✅ Type-safe access to nodes/connections
+- ✅ Better error handling
+- ✅ Direct access to validation results
+- ✅ Easier to debug and test
+
+## API Endpoints (Unchanged)
+
+All endpoints work exactly as before:
+
+### POST /api/llm/chat
+```json
+{
+  "prompt": "What are the main arguments?",
+  "language": "en"
+}
+```
+
+### POST /api/llm/generate-graph
+```json
+{
+  "topic": "Climate change solutions",
+  "language": "en",
+  "max_nodes": 12
+}
+```
+
+### POST /api/llm/review-graph
+```json
+{
+  "language": "en"
+}
+```
+
+## Documentation
+
+- **Full Refactoring Details**: `docs/LLM_REFACTORING.md`
+- **Summary of Changes**: `REFACTORING_SUMMARY.md`
+- **Architecture Overview**: See diagrams in docs
+
+## Need Help?
+
+1. Check `docs/LLM_REFACTORING.md` for detailed architecture
+2. Read `tests/test_llm_refactor.py` for usage examples
+3. All modules have docstrings explaining their purpose
+
+## What's Next?
+
+Consider these enhancements:
+1. Add caching for frequent queries
+2. Enable advanced critique in generation pipeline
+3. Add monitoring for pipeline stages
+4. Implement plugin system for custom backends
+
+See `docs/LLM_REFACTORING.md` section "Future Enhancements" for details.

README.md

@@ -55,11 +55,31 @@ cd thinking_graph
 python -m venv .venv
 source .venv/bin/activate  # Linux/Mac
 # 或 .venv\Scripts\activate  # Windows
+```
+
+#### 选择安装方式
 
-# 安装依赖
+**方式 1：最小运行（API LLM）- 推荐新手**
+```bash
 pip install -r requirements.txt
-# 如果需要本地运行语言模型
-# pip install -r ./requirements-local-llm.txt
+```
+包含 Flask + OpenAI API 客户端，适合使用 DeepSeek/OpenAI/Claude 等云端服务。
+
+**方式 2：本地 LLM/NPU 加速**
+```bash
+pip install -r requirements-local-llm.txt
+```
+包含 ONNX Runtime / OpenVINO，适合本地推理和 NPU 加速。
+
+**方式 3：开发测试环境**
+```bash
+pip install -r requirements-dev.txt
+```
+包含 pytest、httpx 等开发工具。
+
+**方式 4：完整安装（所有依赖）**
+```bash
+pip install -r requirements/all.txt
 ```
 
 ### 配置
@@ -83,6 +103,10 @@ python main.py
 
 打开浏览器访问 `http://localhost:5000`，开始构建你的第一张思维图！
 
+## Production Deployment
+
+For container deployment and multi-user isolation setup, see [DEPLOYMENT.md](DEPLOYMENT.md).
+
 ---
 
 ## 🎯 核心特性