Short answer: in Prefect 3.x, no — “pull steps” don’t exist anymore. Instead, you declare where the code lives when you deploy. Use

Flow.from_source(...).deploy(...)

to point at your repo (or local path) and Prefect will fetch the code at run time. Example:

from prefect.flows import Flow

flow = Flow.from_source(
    source="<https://github.com/your-org/your-repo>",
    entrypoint="flows/etl.py:my_flow",  # path to file + :flow function
)

flow.deploy(
    name="prod",
    work_pool_name="your-pool",
)

- For local code, use

source="."

. - For remote code, use your repo URL. The worker will retrieve the source when it runs the deployment. References - Prefect 3 removed the old build/pull/push steps model; deployments are created directly from source: Prefect 3 release notes - 2.x used pull steps in `prefect.yaml`: Prefect 2 deploy guide - Flow.from_source API (for the curious): flows.py If you’re on Prefect 2.x, pull steps are configured in

prefect.yaml

. If you’re on 3.x (the default assumption), use

from_source

. Which version are you on?

Short version: there isn’t a Python API for “pull steps” in 3.x. Those steps are only honored by

prefect deploy

reading your

prefect.yaml

. With the Python SDK, you specify the code source directly and let the worker fetch it. How to mirror your YAML in Python - Code checkout: point your deployment at the repo with

Flow.from_source(...)

. - Auth: include an access token in the URL (or use env vars/secrets to compose it). - Dependencies: instead of

pip_install_requirements

at runtime, bake deps into the image and pass

image=...

to

deploy

. It’s far more reliable. Example that mirrors your flow:

import os
from prefect.flows import Flow

# Option A (recommended): read a token from env/variable/secret you control
token = os.environ["GITHUB_TOKEN"]
repo = f"<https://x-access-token:{token}@github.com/jane-doe/test-prefect-deployment.git>"

# Load the flow from your repo; entrypoint = "path/to/file.py:flow_function_name"
flow = Flow.from_source(
    source=repo,
    entrypoint="flows/etl.py:my_flow",
)

# Pin dependencies by using a prebuilt image (no pip-at-runtime surprises)
flow.deploy(
    name="prod",
    work_pool_name="your-pool",
    image="<http://ghcr.io/jane-doe/prefect:3.4.8|ghcr.io/jane-doe/prefect:3.4.8>",  # prebuilt image with your requirements
)

Notes - Need to fetch a short-lived token like your

prefect-cloud github token …

step? You can still shell out before calling

from_source

and inject it into the URL, but I’d strongly recommend turning that into a variable/secret so you’re not scripting token fetches at deploy time. - If you must “pin” a specific branch/tag/commit, use a URL that points at the ref (or a storage object that supports refs). If your ref needs more than a simple URL can express, keep using

prefect deploy

with your YAML for that deployment. - The Python SDK path intentionally replaces pull/build/push “steps” with “declare the source + use an image” for reproducibility. References - Prefect 3 changed the deployment model (no build/pull/push pipeline): Prefect 3 release notes - Flow.from_source and Flow.deploy live here if you want to peek: flows.py If you tell me which worker you’re using (process, docker, k8s) and whether you truly need to install requirements at runtime, I can tailor this to your setup. Or, yes, you can keep using the

prefect.yaml

pull steps with `prefect deploy`—they still work in 3.x. I won’t be offended.