Web

Web tools let any agent search the internet or read any webpage, with automatic provider fallback so it works even without an API key.

from praisonaiagents import Agent

agent = Agent(
    name="Researcher",
    instructions="Answer questions using current information from the web.",
    web=True,
)
agent.start("What are the latest developments in AI this week?")

The user asks for current information; the agent searches or fetches pages with automatic provider fallback.

AgentTeam(web=…) is not wired yet (PraisonAI #4004) — set web=… on each Agent(...) in the team instead.

Quick Start

Level 1 — Bool (simplest)

Add web=True and the agent automatically searches the internet when it needs current information.

from praisonaiagents import Agent

agent = Agent(
    name="Researcher",
    instructions="Answer questions using current information from the web.",
    web=True,
)

agent.start("What are the latest developments in AI this week?")

Level 2 — String (pick a provider)

Pass a provider name to route searches through a specific engine.

from praisonaiagents import Agent

agent = Agent(
    name="Researcher",
    instructions="Answer questions using current information from the web.",
    web="tavily",
)

agent.start("What are the latest developments in AI this week?")

Level 3 — Config class (full control)

Use WebConfig to combine search, fetch, provider, and result limits.

from praisonaiagents import Agent, WebConfig

agent = Agent(
    name="Researcher",
    instructions="Research topics thoroughly.",
    web=WebConfig(
        search=True,
        fetch=True,
        search_provider="tavily",
        max_results=10,
    )
)

agent.start("Summarize the latest news on quantum computing.")

Alternative — Web tools directly

Use search_web and web_crawl as standalone tools inside any agent.

from praisonaiagents import Agent
from praisonaiagents.tools import search_web, web_crawl

agent = Agent(
    name="Web Agent",
    instructions="Search and read web pages to answer questions.",
    tools=[search_web, web_crawl],
)

agent.start("Find and summarize the homepage of OpenAI.")

How It Works

The agent decides when to search or fetch based on your query. Results flow back as context for the final answer.

Configuration Options

WebConfig API Reference

Full API reference for WebConfig — search, fetch, providers, max_results

WebConfig Fields

Field	Type	Default	Description
`search`	`bool`	`True`	Enable web search
`fetch`	`bool`	`True`	Enable web page fetching
`search_provider`	`str`	`"duckduckgo"`	Provider to use: `tavily`, `brave`, `exa`, `youdotcom`, `duckduckgo`, `searxng`
`max_results`	`int`	`5`	Maximum search results returned
`search_config`	`dict \| None`	`None`	Provider-specific search settings
`fetch_config`	`dict \| None`	`None`	Provider-specific fetch settings

Search Provider Priority (auto-fallback)

`web_crawl` Output

Every fetched page comes back as a dict with four keys, and the content field is Markdown — headings, lists, links, and fenced code blocks survive even when no optional crawl provider is installed.

Field	Type	Description
`url`	`str`	The URL that was fetched
`content`	`str`	Markdown-formatted page content, capped at 50,000 characters
`title`	`str`	Parsed `<title>` of the page (empty string if missing)
`provider`	`str`	Which backend handled the fetch: `tavily`, `crawl4ai`, `spider`, or `httpx` (built-in, no dependency)

A failed fetch adds an error key instead of usable content.

What survives on the zero-dependency path

The built-in httpx path uses only Python’s standard library, so it works on every install with no API key or extra package.

HTML	Becomes
`<h1>`..`<h6>`	`#` .. `######`
`<ul>` / `<ol>` / `<li>`	`-` list / `1.` ordered list (nested indent aware)
`<a href="...">text</a>`	`[text](url)` (relative URLs resolved against the page)
`<pre><code>...</code></pre>`	Fenced code block (fence grows past inner backticks)
`<code>...</code>` (inline)	`code`
`<strong>` / `<em>`	`bold` / `italic`
`<title>`	Returned in the `title` field
`<script>` / `<style>` / `<head>` / `<noscript>`	Dropped

When an installed provider (tavily, crawl4ai, spider) is available, it is used first for higher-fidelity Markdown. The stdlib httpx path runs as the always-available fallback — and now returns structured Markdown instead of flat text. On any parse error it falls back to flat-text extraction, so the tool never fails.

Example

from praisonaiagents.tools import web_crawl

result = web_crawl("https://example.com/docs/intro")
print(result["title"])     # "Intro — Example Docs"
print(result["provider"])  # "httpx"
print(result["content"][:120])
# # Intro
#
# Welcome to the docs. See [the quick start](https://example.com/docs/quickstart) for a tour.
#
# ```python
# from example import Client
# ```

Common Patterns

Research Agent with Sources

Search the web and return cited sources alongside the answer.

from praisonaiagents import Agent

agent = Agent(
    name="Research Assistant",
    instructions="Answer questions with sources. Always cite URLs.",
    web=True,
)

agent.start("What is the current state of nuclear fusion research?")

Crawl and Summarize a URL

Fetch a specific webpage and summarize it.

from praisonaiagents import Agent
from praisonaiagents.tools import web_crawl

agent = Agent(
    name="Summarizer",
    instructions="Fetch the given URL and write a 3-sentence summary.",
    tools=[web_crawl],
)

agent.start("Summarize https://openai.com/research/overview")

The agent reads the page as Markdown, so headings and code blocks are preserved — the summary can quote them directly.

Multi-Agent Research Pipeline

One agent searches, another writes the report.

from praisonaiagents import Agent, Task, PraisonAIAgents

searcher = Agent(
    name="Searcher",
    instructions="Search the web for information on the given topic.",
    web=True,
)

writer = Agent(
    name="Writer",
    instructions="Write a structured report based on research findings.",
)

search_task = Task(
    description="Find the top 5 sources on AI regulation in 2025",
    agent=searcher,
    expected_output="List of sources with summaries",
)

write_task = Task(
    description="Write a 500-word report on AI regulation",
    agent=writer,
    context=[search_task],
    expected_output="Formatted report",
)

agents = PraisonAIAgents(agents=[searcher, writer], tasks=[search_task, write_task])
agents.start()

Best Practices

Use DuckDuckGo for zero-configuration setups

DuckDuckGo requires no API key and no package beyond ddgs. It’s the best choice for local development, prototypes, and open-source projects where you can’t bundle API keys.

Switch to Tavily for production accuracy

Tavily returns higher-quality snippets and supports full-page extraction. Set TAVILY_API_KEY in your environment and the agent will prefer it automatically over free providers.

Limit max_results to control token costs

Each search result adds tokens to the agent’s context. Keep max_results=5 (the default) for most queries. Only increase it when the task genuinely requires broad coverage, such as academic literature surveys.

Combine search and fetch for deep research

search=True finds the best URLs; fetch=True reads the full page content. Use both together when a snippet is not enough — for example, reading full API documentation or news articles.

Search Providers

Choose DuckDuckGo, Tavily, or Serper for web search

Deep Research

Multi-step research agents that synthesize information from many sources

Getting Started

Install

Core Concepts

Guides

Features

Models

Databases

Observability

Memory

Telemetry & Analytics

Knowledge

RAG

Persistence

Tools

Other Features

Developers

Configuration

Best Practices

Getting Started (No Code)

Quick Start

How It Works

Configuration Options

WebConfig API Reference

WebConfig Fields

Search Provider Priority (auto-fallback)

`web_crawl` Output

What survives on the zero-dependency path

Example

Common Patterns

Research Agent with Sources

Crawl and Summarize a URL

Multi-Agent Research Pipeline

Best Practices

Search Providers

Deep Research

​Quick Start

​How It Works

​Configuration Options

WebConfig API Reference

​WebConfig Fields

​Search Provider Priority (auto-fallback)

​web_crawl Output

​What survives on the zero-dependency path

​Example

​Common Patterns

​Research Agent with Sources

​Crawl and Summarize a URL

​Multi-Agent Research Pipeline

​Best Practices

​Related

Search Providers

Deep Research

Quick Start

How It Works

Configuration Options

WebConfig Fields

Search Provider Priority (auto-fallback)

`web_crawl` Output

What survives on the zero-dependency path

Example

Common Patterns

Research Agent with Sources

Crawl and Summarize a URL

Multi-Agent Research Pipeline

Best Practices

Related