Built-in tools

The tools below are importable from gwenflow.tools. Each one loads its dependency on first use, so importing the package stays light; a tool whose package is missing tells you what to install (the full list is in Installation).

from gwenflow import Agent, ChatOpenAI
from gwenflow.tools import DuckDuckGoSearchTool, WebsiteReaderTool

agent = Agent(
    name="Web researcher",
    llm=ChatOpenAI(model="gpt-5-mini"),
    tools=[DuckDuckGoSearchTool(), WebsiteReaderTool()],
)

Web and research

Tool	Does	Needs
`DuckDuckGoSearchTool`, `DuckDuckGoNewsTool`	web and news search; `region`, `time`, `max_results`, `safesearch`	`duckduckgo-search`
`TavilyWebSearchTool`	web search through Tavily; `max_results`, `search_depth`	`tavily-python`, `TAVILY_API_KEY`
`WebsiteReaderTool`	fetches a URL and returns its text; `max_depth`
`WikipediaTool`	looks up a Wikipedia page	`wikipedia`
`PDFReaderTool`	reads a PDF from a path or a URL	`gwenflow[readers]`
`PubMedTool`	searches PubMed abstracts	`requests`; optional `NCBI_API_KEY`, `NCBI_EMAIL`
`ClinicalTrialsTool`	searches ClinicalTrials.gov	`requests`
`WebSearchTool` (`gwenflow.tools.websearch`)	Google Programmable Search	`WEBSEARCH_API_KEY`, `WEBSEARCH_SEARCH_ENGINE_ID`

Finance

Tool	Does	Needs
`YahooFinanceStock`	price and key figures of a ticker	`yfinance`
`YahooFinanceNews`	latest news of a ticker	`yfinance`
`YahooFinanceScreen`	stock screener	`yfinance`

Files and code

Tool	Does
`ReadFileTool`, `WriteFileTool`, `EditFileTool`	read a file, write one, replace an exact, unique passage
`GrepTool`, `FindTool`, `LsTool`	search contents, find files by name, list a directory
`LocalFileReadTool`, `LocalFileWriteTool`	read and save text files under a `target_directory`
`ShellTool`	runs a shell command in `base_dir`
`PythonCodeTool`	runs Python code in the current environment
`DockerCodeTool`	runs Python or shell code in a throwaway Docker container (`python:3.12-alpine` or `alpine:latest`); needs the `docker` package and a daemon

The file tools take base_dir and, by default, refuse any path outside it (restrict_to_base_dir=True); they also cap what they read (max_lines, max_bytes). They are the bundle of the coding agent.

ShellTool and PythonCodeTool execute whatever the model writes, with the permissions of your process. Use DockerCodeTool for isolation, or declare the tool with requires_approval=True so a human reviews each call. See Tool approvals.

Retrieval

RetrieverTool(retriever=...) lets the model search a knowledge base when it decides to, rather than on every run. See Retrieval (RAG).

Names for flows and the command line

Key	Tool
`wikipedia`	`WikipediaTool`
`duckduckgo`, `duckduckgo-news`	`DuckDuckGoSearchTool`, `DuckDuckGoNewsTool`
`tavily`	`TavilyWebSearchTool`
`website`	`WebsiteReaderTool`
`pdf`	`PDFReaderTool`
`python`	`PythonCodeTool`
`shell`	`ShellTool`
`yahoo-stock`, `yahoo-news`, `yahoo-screen`	`YahooFinanceStock`, `YahooFinanceNews`, `YahooFinanceScreen`

gwenflow tools prints the same list.

Get started

Concepts

Models

Agents

Multi-agent

Extend

Memory

Flows

Knowledge

Production

Project

Web and research

Finance

Files and code

Retrieval

Names for flows and the command line

​Web and research

​Finance

​Files and code

​Retrieval

​Names for flows and the command line

Web and research

Finance

Files and code

Retrieval

Names for flows and the command line