pywebtask 0.20.0

LLM-powered web automation library with autonomous agents

webtask

Easy to use LLM-powered web automation.

Installation

pip install pywebtask
playwright install chromium

Quick Start

from webtask import Webtask
from webtask.integrations.llm import GeminiComputerUse

wt = Webtask()
agent = await wt.create_agent(llm=GeminiComputerUse(), mode="visual")

await agent.do("Go to practicesoftwaretesting.com and add 2 Flat-Head Wood Screws to the cart")

verdict = await agent.verify("the cart contains 2 items")
if verdict:
    print("Success!")

Features

Simple or complex tasks

await agent.do("Click the login button")                            # Single action
await agent.do("Find the blue shirt, add to cart, and checkout")    # Multi-step task

Stateful agents

# Agent remembers context across tasks
await agent.do("Add 2 wood screws to cart")
await agent.do("Add 5 cross-head screws")
await agent.do("Go to cart and verify")

Three modes

agent = await wt.create_agent(llm=llm, mode="text")     # DOM-based (default)
agent = await wt.create_agent(llm=llm, mode="visual")   # Screenshots
agent = await wt.create_agent(llm=llm, mode="full")     # Both

Verification

verdict = await agent.verify("the cart contains 7 items")
if verdict:
    print("Success!")

Data extraction

class ProductInfo(BaseModel):
    name: str
    price: float

product = await agent.extract("product information", ProductInfo)

Error handling

try:
    await agent.do("Add item to cart")
except TaskAbortedError as e:
    print(f"Task failed: {e}")

Easy integration

agent = await wt.create_agent(llm=llm)                                  # New browser
agent = await wt.create_agent_with_browser(llm=llm, browser=browser)    # Existing browser
agent = wt.create_agent_with_context(llm=llm, context=context)          # Existing context
agent = wt.create_agent_with_page(llm=llm, page=page)                   # Existing page

Supported LLMs

from webtask.integrations.llm import Gemini, GeminiComputerUse, Bedrock

Gemini(model="gemini-2.5-flash")                                # 2.5 Flash
GeminiComputerUse(model="gemini-2.5-computer-use-preview")      # Visual mode
Bedrock(model="anthropic.claude-sonnet-4-20250514-v1:0")        # Claude 4 Sonnet (WIP)

TODO

• Unlimited context - Compact conversation history for extended sessions
• Mind2Web benchmark - Evaluation on Mind2Web dataset

Links

• Documentation
• Benchmarks

License

MIT