Run your first workflow locally

Your Union.ai account can display locally run workflows before you connect any cloud infrastructure. A workflow you run with --tracked executes on your own machine and reports its progress to Union.ai as it goes, so you can see how Union.ai works before you create a cluster.

What you’ll need

• A Union.ai organization. See Sign up for Union.ai.
• Your organization’s address, in the form <your-org>.hosted.unionai.cloud. It is in your browser’s address bar when you are signed in to the Union.ai UI.
• Python 3.10+ in a virtual environment.

Set up the CLI

Install the Flyte SDK, which includes the flyte command:

pip install flyte

Create a configuration file that points the CLI at your organization. Replace the endpoint with your organization’s address:

flyte create config \
    --endpoint <your-org>.hosted.unionai.cloud \
    --domain development \
    --project default \
    --builder remote

This writes .flyte/config.yaml in the current directory. The first command that contacts your organization opens a browser window so you can sign in; after that, the CLI remembers you.

Create and run your own workflow

Save this file as hello.py:

import flyte

env = flyte.TaskEnvironment(name="hello_env")

@env.task
def fn(x: int) -> int:
    slope, intercept = 2, 5
    return slope * x + intercept

@env.task
def main(x_list: list[int] = [1, 2, 3, 4, 5]) -> float:
    y_list = list(flyte.map(fn, x_list))
    return sum(y_list) / len(y_list)

Run it with:

flyte run --tracked hello.py main

The --tracked flag runs the workflow on your machine and reports its progress to your organization as it goes.

The example fans a small computation out over a list of inputs with flyte.map and averages the results. The path Union.ai prints is the run’s page in the UI.

See your run in the UI

Open the path printed in your terminal. Or, from your organization’s home page, select Projects, open the default project, and select Tracked Runs. Tracked runs have their own section, separate from Runs, which shows runs that executed on a cluster.

Select main to see the run itself:

The run page shows:

• The run and each of its actions, with status and timing. The example has one parent action and five child actions, one per input.
• The environment the task belongs to.
• Under Summary, the inputs the run received and the outputs it produced.

Everything you see here came from a run on your own machine. Union.ai recorded it as it happened.

Next steps

When you want to start running your workflows on your AWS infrastructure, proceed with:

• Provision your AWS resources. Optional if you already have AWS resources that meet the requirements.
• Connect your cluster.

Before that, two things you can do without a cluster are:

• Try running more code the same way. Write a workflow following the Quickstart. See Track local runs in the console for what tracking does and does not report.
• Learn the concepts. Core concepts explains tasks, environments, projects, and runs.