AGBCLOUD

Appearance

AI Agent operations ​

What you’ll do ​

Control web pages using AGB’s AI Agent with three core operations:

  • Act: perform actions from natural language instructions
  • Observe: find/describe interactable elements
  • Extract: extract structured data using a schema

Prerequisites ​

  • AGB_API_KEY
  • A valid browser image_id (e.g. agb-browser-use-1)
  • A Playwright page connected via CDP (see docs/browser/overview.md)

Quickstart ​

python
from agb.modules.browser import ActOptions

act_result = await session.browser.agent.act_async(
    ActOptions(action="Click the login button"),
    page,
)
print(act_result.success, act_result.message)

Common tasks ​

Act: perform actions ​

python
from agb.modules.browser import ActOptions

await session.browser.agent.act_async(ActOptions(action="Scroll down and click 'Load more'"), page)

Observe: analyze the page ​

python
from agb.modules.browser import ObserveOptions

success, results = await session.browser.agent.observe_async(
    ObserveOptions(instruction="Find all clickable buttons and links on the page"),
    page,
)

if success:
    for r in results:
        print(r.description, r.selector, r.method)

Extract: structured data ​

python
from typing import List
from pydantic import BaseModel
from agb.modules.browser import ExtractOptions


class Product(BaseModel):
    name: str
    price: str
    rating: float
    availability: str


class ProductList(BaseModel):
    products: List[Product]


success, data = await session.browser.agent.extract_async(
    ExtractOptions(
        instruction="Extract all product information from this page",
        schema=ProductList,
        use_text_extract=True,
    ),
    page,
)

if success and data:
    for p in data.products:
        print(p.name, p.price)

Best practices ​

  • Be explicit in instructions (what to click, how to identify it, what to do next).
  • For dynamic pages, ask the agent to wait for content to load before acting.

Troubleshooting ​

Agent actions fail ​

  • Likely cause: ambiguous instructions or page not fully loaded.
  • Fix: add constraints (“button with text…”) and ask the agent to observe page state first.