Short answer: yes — that can be expected. What you’re seeing are “ephemeral” (temporary) Prefect servers that start automatically when you run flows locally without pointing at an existing API. If you kick off multiple runs from separate processes/terminals, each process can start its own ephemeral server. How to avoid multiple ephemeral servers - Use a single persistent local server and point your profile at it:

prefect server start
  # then set your profile to use that API URL so future runs reuse it
  prefect config set PREFECT_API_URL=http://<host>:<port>

Docs: Server CLI - Or disable ephemeral servers entirely (flows will run “offline” and won’t record to a server/UI):

# one-time in your current profile
  prefect config set PREFECT_SERVER__EPHEMERAL__ENABLED=false
  # or via env var
  export PREFECT_SERVER__EPHEMERAL__ENABLED=false

Settings reference: ServerEphemeralSettings Notes - If you’re connected to Prefect Cloud or a self-hosted server (i.e.,

PREFECT_API_URL

is set), Prefect won’t start an ephemeral server. - Ephemeral servers are per-process and shut down when the process exits. If you can share how you’re running flows (e.g.,

python flow.py

,

prefect run

, multiple terminals, etc.), I can suggest the cleanest setup for your case.

Got it — can you share what happens when you disable ephemeral servers? Do you see errors, or are you expecting certain UI/logging features to still work? Here’s what to expect when you truly run “offline” (no server, no ephemeral): - Your flow still runs locally and returns results in-process. - Nothing is recorded to a server/UI: no flow/task states, no logs in the UI, no artifacts, no variables/blocks fetched from API, no automations/concurrency. - Anything that requires the API (e.g., deployments, variables, block documents, artifacts, pause/resume) will fail or be skipped. How to ensure you’re offline (and avoid ephemeral servers) - Disable ephemeral server:

prefect config set PREFECT_SERVER__EPHEMERAL__ENABLED=false

- Make sure you are not pointing at any API:

prefect config unset PREFECT_API_URL

or verify:

prefect config view --show-defaults

You should NOT see

PREFECT_API_URL=http://...

and should NOT see

PREFECT_API_URL=ephemeral

. Optional: silence API logging attempts when offline - Prefect can try to send logs to the API if that feature is on. Turn it off to avoid noise:

prefect config set PREFECT_LOGGING__TO_API__ENABLED=false

Settings ref: Settings reference If you want a single server instead of many ephemeral ones - Start a persistent local server and reuse it:

prefect server start
  prefect config set PREFECT_API_URL=<http://localhost:4200>  # adjust if needed

Docs: Server CLI If you’re seeing different behavior after disabling ephemeral servers, please paste the exact command you’re running and any log output — I’ll help pinpoint what’s still triggering server usage.

Thanks for sharing the settings. That error is expected if your flow (or anything it calls) needs the Prefect API, because you’ve disabled ephemeral servers and haven’t set a persistent

PREFECT_API_URL

. Two important points: - The correct setting to disable ephemeral servers in Prefect 3 is

PREFECT_SERVER__EPHEMERAL__ENABLED=false

(note the double underscores). You can confirm your active values with:

prefect config view --show-defaults

Docs: ServerEphemeralSettings -

PREFECT_SERVER_ALLOW_EPHEMERAL_MODE

is not a Prefect 3 setting; you can remove it. Why you’re seeing the error - With ephemeral disabled, Prefect will not auto-start a server. If your code uses any API-dependent features, you’ll get:

ValueError: No Prefect API URL provided. Please set PREFECT_API_URL ...

- Common API-dependent features: - Deployments (

flow.from_source(...).deploy(...)

,

prefect.deploy

,

prefect.serve

) - Variables or blocks (

Variable.get

,

Block.load

, filesystem blocks) - Artifacts (

create_markdown_artifact

,

create_table_artifact

, etc.) - Client usage (

get_client

,

PrefectClient

) - Pause/resume, automations, concurrency limits, events - Running a deployment by name (

run_deployment

) How to proceed (pick one) 1) Keep running offline (no server) - Remove or guard calls that require the API. - Optionally turn off logging-to-API to avoid noise:

prefect config set PREFECT_LOGGING__TO_API__ENABLED=false

- Then run your flow by calling the function directly (e.g.,

python your_flow.py

). This will work without

PREFECT_API_URL

. 2) Use a persistent local server instead of ephemeral - Start one server and point your profile to it so all runs reuse it:

prefect server start
  prefect config set PREFECT_API_URL=<http://localhost:4200>   # adjust to your host/port

Docs: Server CLI Quick checklist - Verify ephemeral is really off:

prefect config view --show-defaults

Ensure you see

PREFECT_SERVER__EPHEMERAL__ENABLED=false

. - If you want offline: ensure there’s no

PREFECT_API_URL

set and no API-dependent feature is used. - If you want server-backed features: set

PREFECT_API_URL

to a running server. If you paste the stack trace or share which Prefect features your flow uses (e.g., artifacts, variables, deployments), I can point out exactly which call is triggering the API requirement and suggest a workaround.