How hrithikkoduri/webrover works
Project type and purpose
• This is a full-stack AI-powered web automation and research tool called WebRover.
• Purpose: Provide users with a chat-style interface (Next.js frontend) to interact with three specialized agents (Task, Research, Deep Research) hosted in a FastAPI backend. Agents orchestrate browser automation (via Playwright) and LLM-driven workflows (via LangChain, OpenAI, Anthropic, etc.), producing task results, research summaries, or academic-style reports.
File structure
• Top level:
• backend/ – Python FastAPI service, agent orchestrator, browser manager, research modules, marking scripts.
• frontend/ – Next.js + TypeScript UI, styling (Tailwind), components, pages for rover and testing.
• assets/ – media and demo videos.
• Patterns:
• MVC-ish separation: frontend (view+client logic) ↔ backend (controller+model logic).
• Modular agent implementations under backend/app (browser_manager.py, task_agent.py, research_agent.py, deep_research_agent.py).
• LangGraph.json in backend root for state-graph config.
• Config & deployment:
• backend/pyproject.toml, .env expectations, uvicorn launch in main.py.
• frontend/package.json, Next.js config, Tailwind, ESLint, nodemon for local dev.
README for additional insights
• Architecture section highlights four pillars: State Management (LangGraph), Browser Automation (Playwright), Content Processing (RAG + vector store), AI Decision Making (LLMs).