flowco 0.66.0

Flowco a new mixed-initiative system leveraging visual dataflow programming and LLMs to support authoring reliable and robust data analyses.

Flowco

by Emery Berger, Stephen Freund, Eunice Jun, Brooke Simon (ordered alphabetically)

Flowco is a system for authoring data analysis workflows with LLM assistance at every stage of the process. Flowco employs a dataflow programming model that serves as a foundation for reliable LLM-centric programming.

Watch Flowco in Action!

Demo Video	Tutorial Video	Exploratoration	Multiverse Analysis	Logistic Regression

For technical details, see our arXiv paper, Flowco: Rethinking Data Analysis in the Age of LLMs.

Web Service

You can try Flowco on the web here.

[!NOTE] This web service is intended for demonstration and experimentation only. It should scale to a modest number of users, but if it is slow or unresponsive, please try again later or install locally.

Local Installation

Configuration

• Use a conda virtual environment or some other virtual environment.
• Use Python 3.11+.
• Ensure dot is on your path.

OpenAI API Key

[!IMPORTANT]

Flowco needs to be connected to an OpenAI account. Your account will need to have a positive balance for this to work (check your balance). Get a key here.

Once you have an API key, set it as an environment variable called OPENAI_API_KEY.

export OPENAI_API_KEY=<your-api-key>

Installing

From Pypi [Recommended]

With pip:

pip3 install flowco

From Source

Or clone the repo and install as an editable package.

pip3 install -e .

This installs a bunch of normal packages, a custom component for Streamlit, and then Flowco as an editable package that you can run locally (rather than as the web service).

Running

On the command line, run flowco, passing it the directory in which to store its files. That directory should already exist:

mkdir /tmp/example
flowco /tmp/example

The first time you run, it may take 15-20 seconds to bring up the web page. It should launch more quickly after that. Flowco will intially open a welcome.flowco graph. Follow the instructions in the right-hand panel to get started. Then proceed through the numbered tutorials to experiment with additional features.

Use -v to turn on logging.