Workflow Overview

Workflows in AI for Process automate business processes by orchestrating AI models, integrations, and human decision points into a structured, executable sequence. Each workflow run is independent and stateless, producing predictable, repeatable outputs.

Create ──► Configure ──► Deploy ──► Redeploy ──► Undeploy ──► Delete ──► Workflow Management

What is a Workflow

A workflow is a directed graph of nodes connected by transitions. You define the logic—what happens, in what order, under what conditions—and the platform executes it. Workflows do not retain memory between runs; each execution starts fresh from the inputs provided at runtime.

Workflows can:

• Process structured and unstructured data using AI models.
• Integrate with external systems via APIs or pre-built connectors.
• Route execution through conditional branches or parallel paths.
• Pause and wait for human input before resuming.
• Start automatically in response to external events or on a schedule.

The Platform provides the Workflow Builder, a visual no-code or low-code interface for designing workflows by placing and connecting nodes on a canvas. Every element in the workflow is part of a deployable API endpoint after deployment.

Workflow Architecture

Execution model

A workflow runs from a Start node through processing nodes to an End node. Execution follows the connections you define. The platform maintains a context object—a shared runtime state container—for the duration of each run. Nodes read inputs from this context and write their outputs back to it.

[Start Node] ──► [Node A] ──► [Node B] ──► [End Node]
                     │
                     ├──► [Node C]  ─────► [End Node]
                     │    (parallel branch)
                     └──► [Node D]  ─────► [End Node]

Key properties of workflow execution:

• Stateless: the context object is discarded after each run. Workflows don’t persist state across executions.
• Sequential execution: nodes run one after another when tasks depend on prior results.
• Parallel execution: multiple branches from the same node run simultaneously for independent tasks. A single node supports up to 10 outgoing parallel branches.
• Context propagation: nodes reference each other’s outputs using {{context.steps.NodeName.output}} syntax, resolved at runtime without manual wiring.

Start node and workflow initiation

Every workflow begins with a Start node, added automatically when you create a workflow. The Start node is the mandatory entry point and must connect to at least one other node.

The Start node defines:

Setting	Purpose
Input variables	Data passed into the workflow at runtime.
Output variables	Variables captured from the workflow’s results.
Event triggers	Automatic initiation when a specific event occurs in a connected third-party application.
Scheduled execution	Time-based initiation using daily, weekly, monthly, cron, or custom intervals.

Event triggers and schedulers can be used independently or together on the same workflow. When an event trigger fires, event data and metadata are passed into the workflow as variables for downstream nodes to reference. Learn more.

Context object

The context object is the central runtime state container for a workflow execution. It tracks inputs, intermediate outputs, node status, and metadata for all nodes in a single run.

Category	Contents
Run metadata	Identifiers, timestamps, version, and workflow instance information.
Execution state	Each node’s input, output, status, and timing data.
Environment and config	Runtime parameters, environment variables, and execution settings.

Nodes reference data from earlier nodes using {{context.steps.NodeName.output.field}} syntax. This declarative referencing resolves lazily—right before each node executes. Script nodes can also extend the context object dynamically, adding computed values accessible to downstream nodes.

// Reference a previous node's output in any input field, prompt, or condition
{{context.steps.ExtractData.output.invoice_number}}

Nodes

Nodes are the building blocks of a workflow. Each node performs a discrete, self-contained task. Nodes connect through success and failure paths, enabling controlled branching and error handling.

Start Node
  │
  ├──► AI Node ──────────────── calls ──► Model (Model Hub)
  │
  ├──► Integration Node ──────── calls ──► External API / Service
  │
  ├──► Agentic App Node ──────── invokes ──► Deployed AI Agent
  │
  ├──► Function Node  (custom JavaScript or Python logic)
  │
  ├──► Condition Node (branching based on if/else logic)
  │
  ├──► Loop Node     (iterate over a dataset)
  │
  ├──► Human Node   (pause for review or approval)
  │
End Node
        │
        ▼
  Context object (shared across all nodes, discarded after run)

For more information on the different types of nodes, see Node Types.

Create a Workflow

Create a workflow to automate a business process, then configure its runtime behavior, environment variables, and lifecycle settings.

1. Log in to your account.
2. On the Workflows page, select Create new workflow. The New workflow dialog box opens.
3. Enter a name and a brief description for the workflow.
4. Select Create. The workflow is created and the Workflow page opens. You can now build your workflow in the Flow Builder.

To start building workflows by using nodes, see Workflow Builder

---

Import a Workflow

Use the import feature to transfer workflow configurations between environments, create workflows from existing definitions, or apply versioning to existing workflows.

Using the import functionality, you can:

• Create a new workflow. See Import a New Workflow
• Add a new version to an existing workflow. See Import a Workflow as a Version

The import process preserves workflow configuration, including prompts, properties, linked workflows, and external or open-source models, while handling environment variables, errors, and role permissions.

Best Practices

• Export the workflow first: Always export the current workflow version before importing to ensure you have all necessary configurations.
• Check the environment and load dependencies: Before importing, ensure the new environment has all required dependencies—models, linked workflows, and configurations.
  • Verify that the correct model is already available in the new environment.
  • Ensure linked workflows are present in the new environment with the same name for auto-linking.
  • Check environment variables and configuration files to prevent mismatches.
• Import the parent workflow and version: After verifying dependencies, import the parent workflow and any associated versions. Ensure there are no conflicts with existing workflows.

Users with only Viewer permissions cannot import a workflow.

Import a New Workflow

Import a .zip package from your local system containing the flow definition, app definition, and environment variable JSON files from another environment.

Follow these steps to import a workflow and create a new workflow in your account:

1. Log in and click Workflows from the list of modules.
2. Click Import a workflow.
3. In the Import workflow window, provide the following details, and click Import:

Field / Action	Description
Workflow Name	Enter a unique name for the workflow.
Flow definition file	Upload the required flow_definition.json containing canvas definitions and AI node configurations (prompts, hyperparameters, timeout).
Workflow definition file (Optional)	Upload app_definition.json for workflow version information and guardrails.
Workflow environment variable file (Optional)	Upload env_variables.json.

4. After import, you are redirected to the Workflow page.
5. Click Go to Flow to view and access the workflow canvas.

After import:

• The tool is listed under All workflows and My workflows with status In Development.
• Node types, descriptions, input/output variables, scanners, and sync/async setup are preserved.
• If the AI node’s model does not match available models in the current account, the model field is left empty but all prompt definitions are imported.
• API keys are not transferred during import.
• The imported workflow cannot be shared with other users.

Failure scenarios: Import may fail due to an incompatible tool version, corrupt file, internal error, or missing guardrails—triggering an error notification.