Browser Use Agent

This project automates US university applications using Playwright and AI services. Refer to plan.md for the detailed requirements in Chinese.

Mail Receiver

MailReceiver lives under src/modules/mail_receivers/ and exposes multiple backends. Each backend has its own file (e.g. pop3_client.py, tempmail_client.py). Configure the provider and credentials in config/config.yaml and .env.

The knowledge base module uses a Qdrant vector database (see docker-compose.yml).

GPT Browser Handler

The project integrates the browser-use package to control the browser with GPT‑4o. GPTBrowserHandler in src/modules/gpt_browser_handler.py exposes a simple interface:

from modules.gpt_browser_handler import GPTBrowserHandler

handler = GPTBrowserHandler()
handler.run_sync("Open example.edu and take a screenshot")

AI Handler

AIHandler in src/modules/ai_handler.py now uses LiteLLM to access different language models (OpenAI, Claude, Gemini, OpenRouter, DeepSeek, etc.). Methods like generate_text() and generate_embedding() automatically retry on failures.

Orchestrator Memory

Orchestrator maintains an in-memory history of executed steps. Each run generates a small plan using a lightweight model, then executes the steps (open the target URL, ask a larger model to draft an essay, take a screenshot). Models are selected automatically based on task complexity and context size. The context can be retrieved using get_context() for later processing.