Pipelines
A pipeline is Python source that the server parses into a node graph. Each upload of the source produces a new immutable version, identified by a 12-character hash.
Creating and updating
Passing --file (or --source for an inline string) runs the parser and
creates a version. Without either, create makes an empty pipeline.
--tags replaces the existing tags rather than adding to them.
Versions and nodes
version show and node show always emit JSON — they exist to be piped
into something else.
Writing back execution state
node state is the endpoint an execution engine uses to report progress.
--status accepts idle, queued, running, waiting, done, error,
or blocked. Use --error with --status error.
--artifacts takes a JSON object. Three keys are understood:
| Key | Type | Meaning |
|---|---|---|
iframe_url | string | Human review page, rendered in an iframe |
source_url | string | Original data source |
segments_data | any | Segment annotation data |
Invalid JSON is rejected before the request is sent.
Agent workspaces
host uploads a built frontend bundle and returns a page_url to open,
allocating a workspace for it. upload replaces the contents of a workspace
you already have, and optionally takes --source alongside --dist.