Web search

The web search tool queries a search engine, scrapes the top results in parallel and returns a cited answer or the clean page text.

Asset path: aixplain/aixplain-web-search/aixplain · Marketplace card

tip

One call searches and scrapes the top results in parallel. An agent does not need a separate scraper tool or one call per URL.

Try it

from aixplain import Aixplain

aix = Aixplain(api_key="YOUR_API_KEY")

web_search = aix.Tool.get("aixplain/aixplain-web-search/aixplain")
web_search.allowed_actions = ["search"]

agent = aix.Agent(
    name="Web researcher",
    description="Answers questions with live web search and cited sources.",
    instructions="Use web search for anything time-sensitive. Cite your sources.",
    tools=[web_search],
).save()

result = agent.run("What is the latest stable Python release? One sentence.")
print(result.data.output)

Show output

The tool needs no connect step and no config. You get it by path and it is ready to run.

Actions

Action	What it does
search	Searches the web, scrapes the top results and returns an answer with references, or the raw pages

Inputs for search, passed in data:

Parameter	Type	Default	Description
query	str	required	The search query.
result_type	str	"raw"	"raw" returns each page's title, URL, snippet and text. "answer" adds an LLM step that returns a cited answer.
num_results	int	5	How many top results to fetch and scrape.
word_limit	int	300	Maximum words kept per scraped page.
timeout	int	2	Per-URL scrape timeout in seconds. Slower pages are skipped.
max_answer_words	int	none	Caps the answer length in "answer" mode.
llm	str	platform default	Asset ID of the LLM that writes the answer in "answer" mode.
search_asset_id	str	platform default	Asset ID of the underlying search model.
include_domains, exclude_domains	list[str]	none	Restrict results to, or drop results from, these domains. Exclusion wins.
after	str	none	Only results published on or after this year or date, for example "2024-01-31".
gl, hl, location	str	none	Search as if from a country ("us"), in an interface language ("en") or from a place ("Toronto, Ontario, Canada").

Run the tool directly

By default the tool returns the scraped pages. Each record has a title, url, description and the cleaned page text in data.

result = web_search.run(action="search", data={
    "query": "what is retrieval augmented generation",
    "num_results": 3,
    "word_limit": 60,
})
for page in result.data:
    print(page["title"], "|", page["url"])

Show output

Get a cited answer

Set result_type="answer" to add an LLM step. The tool returns a short answer and the URLs that support it.

result = web_search.run(action="search", data={
    "query": "what is retrieval augmented generation",
    "result_type": "answer",
    "num_results": 3,
    "max_answer_words": 60,
})
print(result.data)

Show output

Choose a mode

Mode	Tool calls per question	Who summarizes	Use it when
raw (default)	1	The agent	The agent should read the pages and quote or cite them itself.
answer	1	The tool's LLM	You want a short cited answer with the fewest moving parts.

In raw mode, keep num_results and word_limit modest so large payloads do not fill the agent's context. Scraped text is tag-free, and in answer mode every answer comes with the source URLs that support it.

Clean up

agent.delete()

Next steps

• Tools: see every kind of tool an agent can call.
• Marketplace search: let an agent find models, tools and integrations.
• Agents: configure the agent that calls the tool.
• API reference: tool: every parameter and method.