Great question — it depends on the work pool type you’re deploying to. In Prefect 3.x, there are two common patterns: - Managed work pool (Prefect-managed infrastructure): use a list field named

pip_packages

- Container/pull workers (Docker, Kubernetes, ECS/EC2): set the

EXTRA_PIP_PACKAGES

env var as a space-separated string Examples: 1) Managed work pool (list)

flow.from_source("<https://github.com/org/repo>").deploy(
  name="my-deployment",
  work_pool_name="my-managed-pool",
  job_variables={
    "pip_packages": [
      "pandas==2.2.*",
      "numpy>=1.26",
      "prefect-aws"
    ]
  },
)

2) Container workers (space-separated string via env)

flow.from_source("<https://github.com/org/repo>").deploy(
  name="my-deployment",
  work_pool_name="my-k8s-pool",
  job_variables={
    "env": {
      "EXTRA_PIP_PACKAGES": "pandas==2.2.* numpy>=1.26 prefect-aws"
    }
  },
)

Notes: - You can include version pins and extras (e.g.,

package[extra1,extra2]==1.0.0

). - If you’re using a custom image, it’s best to bake dependencies into the image;

EXTRA_PIP_PACKAGES

is handy for quick additions. Docs: - Customize job variables - Managed work pools - Kubernetes worker example Which work pool type are you using? I can tailor the exact

job_variables

for your setup.