Web Analysis Workflow with Tools, Agents and Structured Output

A complete example showing how to build an autonomous website analysis workflow using Chainless. This example demonstrates tool integration, structured LLM outputs, TaskFlow orchestration and screenshot/HTML extraction using Playwright.

In this example, we build a website analysis pipeline powered by:

Tools for HTML fetching and browser screenshot capturing
An AI Agent that analyzes the website using DeepSeek
A structured output model (WebSiteAnalysisResponse)
A TaskFlow that orchestrates the full workflow

This example replicates a real-world automated web intelligence / site audit workflow.

📦 Project Structure


.
├── main.py
└── types.py   # contains WebSiteAnalysisResponse model

🧩 Response Model (types.py)

This model defines the structured output the agent must return.

from pydantic import BaseModel
from typing import List, Optional


class SocialLinks(BaseModel):
    twitter: Optional[str] = None
    instagram: Optional[str] = None
    email: Optional[str] = None


class SEOInfo(BaseModel):
    canonical_url: Optional[str] = None
    robots: Optional[str] = None
    open_graph: bool = False
    twitter_cards: bool = False
    structured_data: bool = False


class WebAnalysis(BaseModel):
    site_name: Optional[str] = None
    title: Optional[str] = None
    description: Optional[str] = None
    type: Optional[str] = None
    technology: Optional[str] = None
    theme: Optional[str] = None
    key_features: List[str] = []
    social_links: Optional[SocialLinks] = None
    author: Optional[str] = None
    analytics: List[str] = []
    seo: Optional[SEOInfo] = None


class WebSiteAnalysisResponse(BaseModel):
    screenshot_path: Optional[str] = None
    analysis: Optional[WebAnalysis] = None

🧠 Full Example Code (main.py)

from chainless import Tool, Agent, TaskFlow
from chainless.models import ModelNames
from dotenv import load_dotenv

import requests
from playwright.sync_api import sync_playwright

from .types import WebSiteAnalysisResponse

load_dotenv()

# ----------------------------
# Tools
# ----------------------------

def fetch_html(url: str) -> str:
    res = requests.get(url)
    res.raise_for_status()
    return res.text

html_tool = Tool(
    name="HTMLFetcher",
    description="Fetch HTML content from a URL",
    func=fetch_html
)

def take_screenshot(url: str) -> str:
    with sync_playwright() as p:
        browser = p.chromium.launch(headless=True)
        page = browser.new_page()
        page.goto(url, wait_until="networkidle")
        path = "screenshot.png"
        page.screenshot(path=path)
        browser.close()
        return path

screenshot_tool = Tool(
    name="ScreenshotTaker",
    description="Take screenshot of the URL",
    func=take_screenshot
)

# ----------------------------
# Agent
# ----------------------------

agent = Agent(
    name="WebAnalyzerAgent",
    model=ModelNames.DEEPSEEK_CHAT,
    tools=[html_tool, screenshot_tool],
    response_format=WebSiteAnalysisResponse,
    system_prompt=(
        "You are a web analysis assistant.\n"
        "You may call the tools HTMLFetcher and ScreenshotTaker.\n"
        "Extract semantic information from the website.\n"
        "Return structured output strictly following WebSiteAnalysisResponse."
    )
)

# ----------------------------
# TaskFlow
# ----------------------------

flow = TaskFlow(name="SmartWebAnalysis", verbose=True)
flow.add_agent("web_agent", agent)

flow.step("web_agent", {"input": "{{input}}"})

# ----------------------------
# Run the Example
# ----------------------------

if __name__ == "__main__":
    url = "https://www.movixar.com"
    result = flow.run(url)
    print("Final Agent Output:")
    print(result.output)