Overview
Workflows
Sedaro Workflows execute configurable, multi-step processes as containerized steps on Sedaro's hardened, scalable compute infrastructure. A workflow can orchestrate your code, third-party software, and one or many AI agents (Juno) and simulations, alongside data processing and other computational tasks. Steps can run in sequence or in parallel, with dependencies and outputs passed between them.
Run workflows on demand, on a schedule, or in response to artifact changes. Each run records the configuration, resolved content versions, and execution history needed to trace how its outputs were produced.
Instead of manually orchestrating scripts or managing container execution yourself, Workflows handle:
- Scheduling and execution of containerized steps on our hardened, scalable runtime
- Dependency management between steps, passing outputs from one to the next
- Version control integration, recording the definition version and resolved code revisions used by a run
- Runtime source code delivery, so workflows can automatically fetch and clone the exact code they need — from external repositories — directly into the execution environment at runtime, without baking it into a container image
- Runtime configuration, so the same workflow can be run with different inputs without editing any code
- Execution history, retaining each run's pinned content, configuration, status, logs, and outputs
Key Constructs
There are four core constructs in Workflows:
Workflow Definition Template
A Workflow Definition Template is the workflow YAML file that lives in a Git repository for a Git-backed workflow. It is the blueprint — it defines the steps, container images, resource requirements, dependencies, and configurable parameters that make up a workflow. This file is authored and maintained by the workflow developer. A definition can also compose other workflow definitions as steps.
Workflow Definition
A Workflow Definition is a registered workflow in a Project. The current API distinguishes two origins:
git: discovered from a workflow YAML file in a connected, whitelisted repository.assembled: created on the platform by composing existing Git or assembled definitions. Updating an assembled definition publishes a new immutable definition version; it does not rewrite an earlier version.
Use the Workflow Definition API for the origins, version selectors, and composition operations supported by your deployment.
Workflow Instance
A Workflow Instance is a reusable configuration that links to a Workflow Definition. For a Git-backed definition, it selects a specific Git reference through gitReference — a branch, tag, or commit SHA.
- A branch-based instance (e.g.,
main) tracks the branch tip known to the service. - A tag-based instance selects the commit associated with that tag; a tag name should not be treated as an immutable commit identifier.
- A SHA-based instance selects a specific commit.
An assembled instance has no root Git reference and uses its definition's latest published version when resolving content. gitReference is not supported for assembled instances.
Think of an instance as a saved template you can execute many times with different inputs.
Workflow Run
A Workflow Run is a single execution of an instance. At submission, the service pins the root definition and composed members to the resolved versions, applies your runtime configuration, and queues execution. A Git-backed run records gitCommitSha. An assembled run instead records workflowDefinitionVersion, and its root gitCommitSha is null. Status, logs, and outputs are updated as execution proceeds; later definition edits do not change the run's pinned definition content.
Lifecycle of a Workflow
Workflows connect your workflow definitions, your configuration, and scalable runtime infrastructure. For a Git-backed workflow, a Workflow Definition Template — the workflow YAML in your Git repository — describes what should run. The system automatically discovers this template and registers it as a Workflow Definition in your project, keeping it in sync as your Git repository evolves. Assembled definitions are created and versioned on the platform.
From there, you create a Workflow Instance that binds the definition to a specific Git reference — a branch, tag, or commit SHA — for a Git-backed definition, or to the assembled definition's latest published version. The instance is your reusable, named configuration that you can execute as many times as you need with different inputs.
Each time you execute an instance, a Workflow Run is created. The run pins the root definition and composed members at submission, pulls in any required source dependencies during execution preparation, applies your runtime configuration, and executes the pipeline. Consult the Workflow Run API for the record exposed by your deployment.
Reproducibility
Keep these distinct:
- Requested references: branch/tag names in an instance or dependency declaration describe what to resolve; they are not resolved commit SHAs.
- Definition versions: the run pins the Git-backed root commit or immutable assembled definition version, plus the selected versions of composed members. An immutable assembled version may still contain member reference selectors; the run's resolved member versions identify what actually ran.
- Source dependency revisions: external source dependencies are resolved to SHAs during execution preparation after the run leaves the queue, then stored and applied to the executed content. They can resolve later than the root definition.
A null root Git SHA for an assembled run does not mean its members have no Git provenance. Inspect the run's content and composed-step metadata as well as its root version. Version pinning supports traceability, but does not by itself guarantee deterministic results: inputs, container images, external data, and runtime behavior also matter.