An agent-ready browser automation desktop app — a PySide6 + Streamlit front-end built to host an AI agent that sees, reasons, and drives a real Chromium browser over the Chrome DevTools Protocol (CDP).
Knowledge Engine wraps a Streamlit web app inside a native Qt desktop window, giving you a polished, installable UI for an agentic system that is planned to be powered by LangGraph and MCP Playwright.
- 🖥️ Desktop-native UX — PySide6 (
QMainWindow+QWebEngineView) hosting a Streamlit app, with graceful server lifecycle (auto-start, readiness polling, clean shutdown) - 💬 Built-in chat interface — Streamlit chat UI with typed message rendering (text / success / warning / info / error) and persistent session history
- 📊 Status dashboard — live OS info, browser process state, and chat statistics
- 🌐 CDP-ready browser launcher — one-click launch of Chromium with
--remote-debugging-port=9222, PID-tracked with stale-process detection - 🤖 Agent-first backend — FastAPI service exposing LangGraph flows (
/run_chatbot,/run_query) with thread-scoped state - 🔌 MCP Playwright roadmap — designed so the LangGraph agent controls Chromium through MCP Playwright over CDP
┌─────────────────────────────────────────────┐
│ PySide6 Desktop Shell (app.py) │
│ ┌───────────────────────────────────────┐ │
│ │ QWebEngineView ──▶ Streamlit :8501 │ │
│ │ • Chat page (ui.py) │ │
│ │ • Status page (status_page.py) │ │
│ └───────────────────────────────────────┘ │
└──────────────────┬──────────────────────────┘
│ HTTP
┌──────────▼───────────┐
│ FastAPI (main.py) │
│ /run_chatbot │
│ /run_query │
└──────────┬───────────┘
│
┌──────────▼────────────────────┐
│ LangGraph flows (planned) │
│ GraphFlow · RetrivalGraph │
└──────────┬────────────────────┘
│ MCP
┌──────────▼────────────────────┐
│ MCP Playwright ──CDP:9222──▶ │
│ Chromium browser │
└───────────────────────────────┘
| Layer | Technology |
|---|---|
| Desktop shell | PySide6 6.10 (Qt WebEngine) |
| Web UI | Streamlit |
| Agent framework | LangGraph 1.2 |
| LLM tooling | pydantic-ai (OpenAI) |
| Backend API | FastAPI |
| Browser control | Playwright via MCP over CDP |
| Config | python-dotenv |
.
├── app.py # PySide6 shell: launches Streamlit, embeds it in Qt browser
├── entry.py # Streamlit entry point (page navigation)
├── ui.py # Chat page + browser launcher (CDP :9222)
├── status_page.py # Status dashboard
├── main.py # FastAPI service with LangGraph flows
├── graphflow.py # (placeholder) graph flow module
├── requirements.txt # Python dependencies
└── .env # Environment variables (not tracked)
The agent backend package (
src/graphflow) is a work-in-progress and intentionally excluded from this repository via.gitignore.
- Python 3.10+
- Chromium (Linux) or Google Chrome (macOS) installed and on
PATH
git clone <repo-url>
cd "Auto Browser"
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt# Launch the desktop app (starts Streamlit automatically on :8501)
python app.py
# Or run the Streamlit UI directly
streamlit run entry.py --server.port 8501In the Chat page, click "Check OS and Launch Browser" in the sidebar to start Chromium with remote debugging enabled on port 9222.
The FastAPI service (once the agent graphs are wired in) exposes:
| Endpoint | Body | Description |
|---|---|---|
POST /run_chatbot |
{"query": "...", "thread_id": "..."} |
Run the chatbot graph flow |
POST /run_query |
{"query": "...", "thread_id": "..."} |
Run the retrieval graph flow |
- Implement
src/graphflow— LangGraphGraphFlowandRetrivalGraphcompilation - Connect the LangGraph agent to Playwright MCP server
- Drive Chromium via CDP (
ws://localhost:9222) for browser automation tasks - Surface agent actions / observations in the chat UI
- Windows support for browser launch
Distributed under the MIT License. See the LICENSE file for details.