/google-antigravity-sdk

Web Tools Examples

This guide demonstrates how to enable and use built-in web search and URL content reading tools with an agent.

Enabling Web Search (search_web)

To use the web search tool, you must explicitly enable SEARCH_WEB in the CapabilitiesConfig passed to LocalAgentConfig.

from google.antigravity import Agent, LocalAgentConfig, CapabilitiesConfig, types

# Configure the agent to enable the SEARCH_WEB tool.
config = LocalAgentConfig(
    capabilities=CapabilitiesConfig(
        enabled_tools=[
            types.BuiltinTools.SEARCH_WEB,
        ]
    ),
)

async with Agent(config) as agent:
    response = await agent.chat("What is the current weather and temperature in New York City right now? Please provide the source.")
    print(await response.text())

[!NOTE] When SEARCH_WEB is enabled, the agent can perform web searches to ground its responses with up-to-date information.

Reading URL Content (read_url_content)

To fetch and extract page content directly from a URL, enable READ_URL_CONTENT:

# Configure the agent to enable URL content reading.
config = LocalAgentConfig(
    capabilities=CapabilitiesConfig(
        enabled_tools=[
            types.BuiltinTools.READ_URL_CONTENT,
            # Note: For massive web pages or lengthy documentation,
            # read_url_content caches the downloaded payload to disk.
            # Enabling VIEW_FILE allows the agent to inspect those files.
            types.BuiltinTools.VIEW_FILE,
        ]
    ),
)

async with Agent(config) as agent:
    target_url = "https://en.wikipedia.org/wiki/Google"
    prompt = f"Please read the full page content from {target_url} and tell me the exact date that Google acquired DeepMind Technologies."
    response = await agent.chat(prompt)
    print(await response.text())

[!TIP] read_url_content scrapes static HTML pages into structured markdown chunks, allowing the model to analyze long web articles or documentation without manual copying. For large pages that exceed inline context thresholds, the payload is cached to disk; pairing READ_URL_CONTENT with VIEW_FILE ensures the agent can read those cached files.