Command line

Installing the library also installs the gwenflow command. It is a thin layer over the library for quick experiments and scripted flow runs.

gwenflow --help
gwenflow --version

Models and tools by name

The command line names models as provider/model-name and tools by a short key, the same names flow files use.

Provider key	Client
`openai`	`ChatOpenAI`
`anthropic`	`ChatAnthropic`
`google`, `gemini`	`ChatGoogle`
`mistral`	`ChatMistral`
`azure`	`ChatAzureOpenAI`
`deepseek`	`ChatDeepSeek`
`ollama`	`ChatOllama`
`gwenlake`	`ChatGwenlake`

A bare model name (gpt-5-mini) means OpenAI. The default model is openai/gpt-4o-mini. The provider’s API key comes from the environment, as in Python (see Models). gwenflow tools lists the tool keys:

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`

The same mapping is available in Python, in gwenflow.registry:

from gwenflow.registry import make_llm, make_tools

llm = make_llm("anthropic/claude-sonnet-4-6", temperature=0)
tools = make_tools(["wikipedia", "duckduckgo"])

`ask`: one question to one agent

gwenflow ask "Summarize the history of the Eiffel Tower in three bullet points." \
    --model anthropic/claude-sonnet-4-6 \
    --tools wikipedia

Option	Meaning
`--model`, `-m`	the model, `provider/model-name`
`--tools`, `-t`	a built-in tool to give the agent; repeat it for several
`--instructions`, `-i`	the agent’s instructions (default: `You are a helpful assistant.`)

The answer is rendered as Markdown in the terminal.

`run`: run a flow file

gwenflow run flows/news_summary.yaml

Runs a flow defined in YAML or JSON and prints the result of its output nodes (the nodes no other node consumes). A failing node prints its error and exits with status 1.

Option	Meaning
`--plot`	write a static PNG of the graph instead of running it (needs `pydot`)
`--interactive`	open an interactive HTML graph instead of running it (needs `pyvis`)

`autoflow`: plan and run a team

gwenflow autoflow "Write a short biography of Ada Lovelace, then list five related topics." \
    --tools wikipedia

Asks the model to break the query into agents with dependencies, then runs them as a flow. See AutoFlow.

Option	Meaning
`--model`, `-m`	the model that plans and runs the agents
`--tools`, `-t`	the tools the planned agents may use; repeatable

Get started

Concepts

Models

Agents

Multi-agent

Extend

Memory

Flows

Knowledge

Production

Project

Models and tools by name

`ask`: one question to one agent

`run`: run a flow file

`autoflow`: plan and run a team

​Models and tools by name

​ask: one question to one agent

​run: run a flow file

​autoflow: plan and run a team

Models and tools by name

`ask`: one question to one agent

`run`: run a flow file

`autoflow`: plan and run a team