Great question. In Prefect 2.20.11 you can create/register a deployment entirely from Python by constructing a

Deployment

and calling

apply()

. Here are the most common patterns, from minimal to more “real world.” Minimal (local code, default Process infrastructure, default work queue)

from prefect import flow
from prefect.deployments import Deployment

@flow
def my_flow():
    return "hi"

# Best in 2.x: build from the flow object so entrypoint is inferred
deployment = Deployment.build_from_flow(
    flow=my_flow,
    name="my-deployment",
    work_queue_name="default",  # or your queue
)

deployment.apply()  # registers/updates the deployment

Agent model: custom infrastructure block (e.g., Docker) + work queue

from prefect import flow
from prefect.deployments import Deployment
from prefect.infrastructure import DockerContainer

@flow
def my_flow():
    ...

docker = DockerContainer(
    image="my-registry/my-image:latest",
    image_pull_policy="ALWAYS",
    env={"PREFECT_LOGGING_LEVEL": "INFO"},
)

deployment = Deployment.build_from_flow(
    flow=my_flow,
    name="docker-deploy",
    work_queue_name="prod",   # agents poll queues
    infrastructure=docker,    # infra block is used with agents
)

deployment.apply()

Worker model: target a work pool (no infra block; use job_variables)

from prefect import flow
from prefect.deployments import Deployment

@flow
def my_flow():
    ...

deployment = Deployment.build_from_flow(
    flow=my_flow,
    name="pool-deploy",
    work_pool_name="k8s-prod",  # workers pull from pools
    # Customize the worker’s base job template via job_variables
    job_variables={
        "env": {"PREFECT_LOGGING_LEVEL": "DEBUG"},
        # other fields match your pool’s base job template
    },
)

deployment.apply()

Remote storage (S3) with upload

from prefect import flow
from prefect.deployments import Deployment
from prefect.filesystems import S3

@flow
def my_flow():
    ...

storage = S3.load("my-s3-block")  # created beforehand in UI/API

deployment = Deployment.build_from_flow(
    flow=my_flow,
    name="s3-deploy",
    storage=storage,
    path="flows",                 # path within the bucket
)

deployment.apply(upload=True)     # uploads the project files to S3

Scheduling (2.x supports either a single schedule or multiple schedules; multiple is preferred)

from prefect.client.schemas.schedules import CronSchedule
from prefect.client.schemas.objects import MinimalDeploymentSchedule

deployment = Deployment.build_from_flow(
    flow=my_flow,
    name="scheduled-deploy",
    schedules=[
        MinimalDeploymentSchedule(
            schedule=CronSchedule(cron="0 2 * * *"),
            active=True,
        )
    ],
)

deployment.apply()

What’s required for apply() - At minimum, a deployment

name

and a flow target. - If you use

Deployment.build_from_flow(...)

, the flow target is inferred and the entrypoint is auto-generated. - If you instantiate

Deployment(...)

manually instead, you must provide

name

,

flow_name

, and typically

entrypoint

and

path

if Prefect can’t infer them. - Optional but common: - Agent model:

work_queue_name

and an

infrastructure

block - Worker model:

work_pool_name

and

job_variables

-

storage

for remote code plus

upload=True

to push files Notes and tips - In CI (e.g., Concourse), ensure

PREFECT_API_URL

(and

PREFECT_API_KEY

for Cloud) are set before running this Python code so

apply()

can register with your server. -

apply(upload=True)

only uploads when you’ve configured a remote storage block. - In 2.x,

Deployment.build_from_flow(...).apply()

is the cleanest route;

Deployment.apply()

is deprecated in favor of 3.x patterns but is still supported in 2.20.11. - When you move to 3.x, switch to

flow.deploy(...)

or

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

and the

prefect deploy

CLI;

Deployment.build_from_flow()

is removed in 3.x.

If you share whether you’re using agents + work queues or workers + work pools (and whether your code lives in remote storage vs. the runtime workspace), I can tailor the snippet exactly to your setup. Also, if you’re hitting failures in Concourse, drop the error message and I’ll help pinpoint the cause.