Lifecycle
new
save
deploy=True vs deploy=False — the branch model
Think of a pipeline like a Git repo:
- Working tree — the
Pipelineobject in your Python process. Everypipeline.add(...)mutation lives here. - Main branch — the
branch_idon the server.save(deploy=False)writes your working tree to this branch as a new commit. Anything reading bybranch_id(the platform editor, ad-hocpipeline.run(...)from this SDK process) immediately sees the change. - Deployed version — what callers of the published pipeline get. Chatbots, integrations, scheduled runs, and the platform’s “Run published” button all read this version.
save(deploy=True)(the default) promotes the current main-branch state to be the new deployed version.
bump attaches a semantic-version tag ("major", "minor", or "patch") to the deploy. Consumers can then pin to that version via Pipeline.fetch(id=..., version="1.2.0") instead of always tracking latest.
In practice:
- Iterating on a draft?
save(deploy=False)— keeps the deployed version stable while you tweak. - Ready to ship?
save(deploy=True)(default) — optionally withbump="patch"+description="..."for a labelled release.
When
True, promotes the saved state to be the new deployed version (the one chatbots, integrations, and the “Run published” path will hit). When False, only updates the working/main branch — the deployed version is untouched until the next save(deploy=True).Semantic-version bump level:
"major", "minor", or "patch". Only meaningful with deploy=True. See BumpLevel.Updates the pipeline description; also used as the changelog entry when bumping a version.
fetch
branch_id and version only apply when fetching by id.
Parameters
The name of the pipeline to fetch. When multiple pipelines share the same name, the most recently modified one is returned.
list
Include shared pipelines.
delete
duplicate
Building
add
add_batch
execution_mode='batch'.
Usage
Node catalogue
pipeline.add(name=).<node_type>(...) exposes every node type as a typed builder method. There are ~500 node types in total. The biggest categories:
Each builder returns a typed node object with attributes (
.text, .response, .results, .path_0, etc.) you wire into downstream nodes. The full surface ships as a .pyi stub (vectorshift/pipeline/node_adder.pyi) — your editor’s autocomplete is the canonical browseable catalogue.
add_node
remove_node
Running
run
Dictionary of input nodes -> input values for the pipeline. eg: {“input_node”: “Hello, world!”}
Whether to stream the response. (Set true only when pipeline has an output node with a streaming llm input)
Optional session ID for run grouping and tracing. Groups multiple runs under the same session for analytics and observability. Note: this does not provide conversational memory to LLM nodes — each run’s LLM nodes only see the current inputs.
Union[dict[str, Any], Generator]: A dictionary containing pipeline outputs and run_id. If stream is True, returns a generator that yields response chunks.
bulk_run
A single dictionary with two keys:
run_outputs — a list with one entry per
input set, each containing that run’s outputs — and status, the overall
bulk-run status. (Note: this is one dict wrapping all runs, not a list.)run_status
A :class:
RunStatus dict with keys task_id, status, and
optionally error or result. See RunStatus.Background runs
start
Optional session ID for run grouping and tracing. Groups multiple runs under the same session for analytics and observability. Note: this does not provide conversational memory to LLM nodes — each run’s LLM nodes only see the current inputs.
terminate
Streaming
stream
Optional session ID for run grouping and tracing. Groups multiple runs under the same session for analytics and observability. Note: this does not provide conversational memory to LLM nodes — each run’s LLM nodes only see the current inputs.
Sharing & publishing
share
unshare
publish
unpublish
move_to_folder
Versioning
revert
Serialization
to_dict
from_json
serialize_inputs
Types
Configuration objects, response shapes, and enums used by the methods above.BumpLevel
Members
PATCH="patch"MINOR="minor"MAJOR="major"
RunResult
Response from :meth:Pipeline.run / :meth:Pipeline.arun.
Fields
RunStatus
Response from :meth:Pipeline.run_status / :meth:RunHandler.run_status.
Fields
StartResult
Response from :meth:Pipeline.start background run initiation.
Fields
