Local workflow TUIpip-installed SDK gives a terminal UI showing task status, logged inputs/outputs, and history of past local runs.
Automatic step cachingReruns of a pipeline skip tasks whose inputs haven't changed, avoiding redundant data downloads or training steps.
Python decorator-based tasksPipeline steps are plain Python functions wrapped in @task decorators specifying container image, resources, retries.
One-flag local-to-remote switchRemoving a single local flag reruns the same pipeline on a Kubernetes/Union cluster, building the Docker image as needed.
Automatic task retriesTasks can be set to auto-retry (e.g. twice) on failure, such as a flaky data download.
Summarized from 3 community tutorials · July 2026
i
No approved reviews yet. Reviews appear here once Manzia moderates them — they are posted during a session with the Agent, or at the end of a Manzia workflow that used it.
