ai-agent-book 精选快照（<2MB 代码与文档，来自 github.com/bojieli/ai-agent-book）

chapter9/browser-use-rpa/browser-use/examples/use-cases/captcha.py

@@ -0,0 +1,35 @@
+"""
+Goal: Automates CAPTCHA solving on a demo website.
+
+
+Simple try of the agent.
+@dev You need to add OPENAI_API_KEY to your environment variables.
+NOTE: captchas are hard. For this example it works. But e.g. for iframes it does not.
+for this example it helps to zoom in.
+"""
+
+import asyncio
+import os
+import sys
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+from browser_use import Agent, ChatOpenAI
+
+
+async def main():
+	llm = ChatOpenAI(model='gpt-4.1-mini')
+	agent = Agent(
+		task='go to https://captcha.com/demos/features/captcha-demo.aspx and solve the captcha',
+		llm=llm,
+	)
+	await agent.run()
+	input('Press Enter to exit')
+
+
+if __name__ == '__main__':
+	asyncio.run(main())

chapter9/browser-use-rpa/browser-use/examples/use-cases/check_appointment.py

@@ -0,0 +1,52 @@
+# Goal: Checks for available visa appointment slots on the Greece MFA website.
+
+import asyncio
+import os
+import sys
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+from pydantic import BaseModel
+
+from browser_use import ChatOpenAI
+from browser_use.agent.service import Agent
+from browser_use.tools.service import Tools
+
+if not os.getenv('OPENAI_API_KEY'):
+	raise ValueError('OPENAI_API_KEY is not set. Please add it to your environment variables.')
+
+tools = Tools()
+
+
+class WebpageInfo(BaseModel):
+	"""Model for webpage link."""
+
+	link: str = 'https://appointment.mfa.gr/en/reservations/aero/ireland-grcon-dub/'
+
+
+@tools.action('Go to the webpage', param_model=WebpageInfo)
+def go_to_webpage(webpage_info: WebpageInfo):
+	"""Returns the webpage link."""
+	return webpage_info.link
+
+
+async def main():
+	"""Main function to execute the agent task."""
+	task = (
+		'Go to the Greece MFA webpage via the link I provided you.'
+		'Check the visa appointment dates. If there is no available date in this month, check the next month.'
+		'If there is no available date in both months, tell me there is no available date.'
+	)
+
+	model = ChatOpenAI(model='gpt-4.1-mini')
+	agent = Agent(task, model, tools=tools, use_vision=True)
+
+	await agent.run()
+
+
+if __name__ == '__main__':
+	asyncio.run(main())

chapter9/browser-use-rpa/browser-use/examples/use-cases/extract_pdf_content.py

@@ -0,0 +1,38 @@
+#!/usr/bin/env -S uv run --script
+# /// script
+# requires-python = ">=3.11"
+# dependencies = ["browser-use", "mistralai"]
+# ///
+
+import os
+import sys
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+import asyncio
+import logging
+
+from browser_use import Agent, ChatOpenAI
+
+logger = logging.getLogger(__name__)
+
+
+async def main():
+	agent = Agent(
+		task="""
+        Objective: Navigate to the following UR, what is on page 3?
+
+        URL: https://docs.house.gov/meetings/GO/GO00/20220929/115171/HHRG-117-GO00-20220929-SD010.pdf
+        """,
+		llm=ChatOpenAI(model='gpt-4.1-mini'),
+	)
+	result = await agent.run()
+	logger.info(result)
+
+
+if __name__ == '__main__':
+	asyncio.run(main())

chapter9/browser-use-rpa/browser-use/examples/use-cases/find_and_apply_to_jobs.py

@@ -0,0 +1,157 @@
+"""
+Goal: Searches for job listings, evaluates relevance based on a CV, and applies
+
+@dev You need to add OPENAI_API_KEY to your environment variables.
+Also you have to install PyPDF2 to read pdf files: pip install PyPDF2
+"""
+
+import asyncio
+import csv
+import logging
+import os
+import sys
+from pathlib import Path
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+from pydantic import BaseModel
+from PyPDF2 import PdfReader  # type: ignore
+
+from browser_use import ActionResult, Agent, ChatOpenAI, Tools
+from browser_use.browser import BrowserProfile, BrowserSession
+
+required_env_vars = ['AZURE_OPENAI_KEY', 'AZURE_OPENAI_ENDPOINT']
+for var in required_env_vars:
+	if not os.getenv(var):
+		raise ValueError(f'{var} is not set. Please add it to your environment variables.')
+
+logger = logging.getLogger(__name__)
+# full screen mode
+tools = Tools()
+
+# NOTE: This is the path to your cv file
+# create a dummy cv
+CV = Path.cwd() / 'dummy_cv.pdf'
+with open(CV, 'w') as f:
+	f.write('Hi I am a machine learning engineer with 3 years of experience in the field')
+
+logger.info(f'Using dummy cv at {CV}')
+
+
+class Job(BaseModel):
+	title: str
+	link: str
+	company: str
+	fit_score: float
+	location: str | None = None
+	salary: str | None = None
+
+
+@tools.action('Save jobs to file - with a score how well it fits to my profile', param_model=Job)
+def save_jobs(job: Job):
+	with open('jobs.csv', 'a', newline='') as f:
+		writer = csv.writer(f)
+		writer.writerow([job.title, job.company, job.link, job.salary, job.location])
+
+	return 'Saved job to file'
+
+
+@tools.action('Read jobs from file')
+def read_jobs():
+	with open('jobs.csv') as f:
+		return f.read()
+
+
+@tools.action('Read my cv for context to fill forms')
+def read_cv():
+	pdf = PdfReader(CV)
+	text = ''
+	for page in pdf.pages:
+		text += page.extract_text() or ''
+	logger.info(f'Read cv with {len(text)} characters')
+	return ActionResult(extracted_content=text, include_in_memory=True)
+
+
+@tools.action(
+	'Upload cv to element - call this function to upload if element is not found, try with different index of the same upload element',
+)
+async def upload_cv(index: int, browser_session: BrowserSession):
+	path = str(CV.absolute())
+
+	# Get the element by index
+	dom_element = await browser_session.get_element_by_index(index)
+
+	if dom_element is None:
+		logger.info(f'No element found at index {index}')
+		return ActionResult(error=f'No element found at index {index}')
+
+	# Check if it's a file input element
+	if not browser_session.is_file_input(dom_element):
+		logger.info(f'Element at index {index} is not a file upload element')
+		return ActionResult(error=f'Element at index {index} is not a file upload element')
+
+	try:
+		# Dispatch upload file event with the file input element
+		from browser_use.browser.events import UploadFileEvent
+
+		event = browser_session.event_bus.dispatch(UploadFileEvent(node=dom_element, file_path=path))
+		await event
+		await event.event_result(raise_if_any=True, raise_if_none=False)
+		msg = f'Successfully uploaded file "{path}" to index {index}'
+		logger.info(msg)
+		return ActionResult(extracted_content=msg)
+	except Exception as e:
+		logger.debug(f'Error in upload: {str(e)}')
+		return ActionResult(error=f'Failed to upload file to index {index}')
+
+
+browser_session = BrowserSession(
+	browser_profile=BrowserProfile(
+		executable_path='/Applications/Google Chrome.app/Contents/MacOS/Google Chrome',
+		disable_security=True,
+		user_data_dir='~/.config/browseruse/profiles/default',
+	)
+)
+
+
+async def main():
+	# ground_task = (
+	# 	'You are a professional job finder. '
+	# 	'1. Read my cv with read_cv'
+	# 	'2. Read the saved jobs file '
+	# 	'3. start applying to the first link of Amazon '
+	# 	'You can navigate through pages e.g. by scrolling '
+	# 	'Make sure to be on the english version of the page'
+	# )
+	ground_task = (
+		'You are a professional job finder. '
+		'1. Read my cv with read_cv'
+		'find ml internships in and save them to a file'
+		'search at company:'
+	)
+	tasks = [
+		ground_task + '\n' + 'Google',
+		# ground_task + '\n' + 'Amazon',
+		# ground_task + '\n' + 'Apple',
+		# ground_task + '\n' + 'Microsoft',
+		# ground_task
+		# + '\n'
+		# + 'go to https://nvidia.wd5.myworkdayjobs.com/en-US/NVIDIAExternalCareerSite/job/Taiwan%2C-Remote/Fulfillment-Analyst---New-College-Graduate-2025_JR1988949/apply/autofillWithResume?workerSubType=0c40f6bd1d8f10adf6dae42e46d44a17&workerSubType=ab40a98049581037a3ada55b087049b7 NVIDIA',
+		# ground_task + '\n' + 'Meta',
+	]
+	model = ChatOpenAI(model='gpt-4.1-mini')
+
+	agents = []
+	for task in tasks:
+		agent = Agent(task=task, llm=model, tools=tools, browser_session=browser_session)
+		agents.append(agent)
+
+	await asyncio.gather(*[agent.run() for agent in agents])
+
+
+if __name__ == '__main__':
+	asyncio.run(main())

chapter9/browser-use-rpa/browser-use/examples/use-cases/find_influencer_profiles.py

@@ -0,0 +1,89 @@
+"""
+Show how to use custom outputs.
+
+@dev You need to add OPENAI_API_KEY to your environment variables.
+"""
+
+import asyncio
+import json
+import os
+import sys
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+import httpx
+from pydantic import BaseModel
+
+from browser_use import Agent, ChatOpenAI, Tools
+from browser_use.agent.views import ActionResult
+
+
+class Profile(BaseModel):
+	platform: str
+	profile_url: str
+
+
+class Profiles(BaseModel):
+	profiles: list[Profile]
+
+
+tools = Tools(exclude_actions=['search_google'], output_model=Profiles)
+BEARER_TOKEN = os.getenv('BEARER_TOKEN')
+
+if not BEARER_TOKEN:
+	# use the api key for ask tessa
+	# you can also use other apis like exa, xAI, perplexity, etc.
+	raise ValueError('BEARER_TOKEN is not set - go to https://www.heytessa.ai/ and create an api key')
+
+
+@tools.registry.action('Search the web for a specific query')
+async def search_web(query: str):
+	keys_to_use = ['url', 'title', 'content', 'author', 'score']
+	headers = {'Authorization': f'Bearer {BEARER_TOKEN}'}
+	async with httpx.AsyncClient() as client:
+		response = await client.post(
+			'https://asktessa.ai/api/search',
+			headers=headers,
+			json={'query': query},
+		)
+
+	final_results = [
+		{key: source[key] for key in keys_to_use if key in source}
+		for source in await response.json()['sources']
+		if source['score'] >= 0.2
+	]
+	# print(json.dumps(final_results, indent=4))
+	result_text = json.dumps(final_results, indent=4)
+	print(result_text)
+	return ActionResult(extracted_content=result_text, include_in_memory=True)
+
+
+async def main():
+	task = (
+		'Go to this tiktok video url, open it and extract the @username from the resulting url. Then do a websearch for this username to find all his social media profiles. Return me the links to the social media profiles with the platform name.'
+		' https://www.tiktokv.com/share/video/7470981717659110678/  '
+	)
+	model = ChatOpenAI(model='gpt-4.1-mini')
+	agent = Agent(task=task, llm=model, tools=tools)
+
+	history = await agent.run()
+
+	result = history.final_result()
+	if result:
+		parsed: Profiles = Profiles.model_validate_json(result)
+
+		for profile in parsed.profiles:
+			print('\n--------------------------------')
+			print(f'Platform:         {profile.platform}')
+			print(f'Profile URL:      {profile.profile_url}')
+
+	else:
+		print('No result')
+
+
+if __name__ == '__main__':
+	asyncio.run(main())

chapter9/browser-use-rpa/browser-use/examples/use-cases/shopping.py

@@ -0,0 +1,120 @@
+import asyncio
+import os
+import sys
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+from browser_use import Agent, ChatOpenAI
+
+task = """
+   ### Prompt for Shopping Agent – Migros Online Grocery Order
+
+**Objective:**
+Visit [Migros Online](https://www.migros.ch/en), search for the required grocery items, add them to the cart, select an appropriate delivery window, and complete the checkout process using TWINT.
+
+**Important:**
+- Make sure that you don't buy more than it's needed for each article.
+- After your search, if you click  the "+" button, it adds the item to the basket.
+- if you open the basket sidewindow menu, you can close it by clicking the X button on the top right. This will help you navigate easier.
+---
+
+### Step 1: Navigate to the Website
+- Open [Migros Online](https://www.migros.ch/en).
+- You should be logged in as Nikolaos Kaliorakis
+
+---
+
+### Step 2: Add Items to the Basket
+
+#### Shopping List:
+
+**Meat & Dairy:**
+- Beef Minced meat (1 kg)
+- Gruyère cheese (grated preferably)
+- 2 liters full-fat milk
+- Butter (cheapest available)
+
+**Vegetables:**
+- Carrots (1kg pack)
+- Celery
+- Leeks (1 piece)
+- 1 kg potatoes
+
+At this stage, check the basket on the top right (indicates the price) and check if you bought the right items.
+
+**Fruits:**
+- 2 lemons
+- Oranges (for snacking)
+
+**Pantry Items:**
+- Lasagna sheets
+- Tahini
+- Tomato paste (below CHF2)
+- Black pepper refill (not with the mill)
+- 2x 1L Oatly Barista(oat milk)
+- 1 pack of eggs (10 egg package)
+
+#### Ingredients I already have (DO NOT purchase):
+- Olive oil, garlic, canned tomatoes, dried oregano, bay leaves, salt, chili flakes, flour, nutmeg, cumin.
+
+---
+
+### Step 3: Handling Unavailable Items
+- If an item is **out of stock**, find the best alternative.
+- Use the following recipe contexts to choose substitutions:
+  - **Pasta Bolognese & Lasagna:** Minced meat, tomato paste, lasagna sheets, milk (for béchamel), Gruyère cheese.
+  - **Hummus:** Tahini, chickpeas, lemon juice, olive oil.
+  - **Chickpea Curry Soup:** Chickpeas, leeks, curry, lemons.
+  - **Crispy Slow-Cooked Pork Belly with Vegetables:** Potatoes, butter.
+- Example substitutions:
+  - If Gruyère cheese is unavailable, select another semi-hard cheese.
+  - If Tahini is unavailable, a sesame-based alternative may work.
+
+---
+
+### Step 4: Adjusting for Minimum Order Requirement
+- If the total order **is below CHF 99**, add **a liquid soap refill** to reach the minimum. If it;s still you can buy some bread, dark chockolate.
+- At this step, check if you have bought MORE items than needed. If the price is more then CHF200, you MUST remove items.
+- If an item is not available, choose an alternative.
+- if an age verification is needed, remove alcoholic products, we haven't verified yet.
+
+---
+
+### Step 5: Select Delivery Window
+- Choose a **delivery window within the current week**. It's ok to pay up to CHF2 for the window selection.
+- Preferably select a slot within the workweek.
+
+---
+
+### Step 6: Checkout
+- Proceed to checkout.
+- Select **TWINT** as the payment method.
+- Check out.
+- 
+- if it's needed the username is: nikoskalio.dev@gmail.com 
+- and the password is : TheCircuit.Migros.dev!
+---
+
+### Step 7: Confirm Order & Output Summary
+- Once the order is placed, output a summary including:
+  - **Final list of items purchased** (including any substitutions).
+  - **Total cost**.
+  - **Chosen delivery time**.
+
+**Important:** Ensure efficiency and accuracy throughout the process."""
+
+
+agent = Agent(task=task, llm=ChatOpenAI(model='gpt-4.1-mini'))
+
+
+async def main():
+	await agent.run()
+	input('Press Enter to close the browser...')
+
+
+if __name__ == '__main__':
+	asyncio.run(main())