Skip to main content

What Are Workflows?

Workflows are visual, node-based systems that control how traces are processed. You build workflows by connecting nodes in a canvas - triggers, conditions, human intervention gates, notifications, and rejections. When a trace comes in and matches a trigger, the workflow executes node by node until it reaches an end state.

How Workflows Work

Trace Created

A trace arrives from your AI system

Policy Assessment

Policies evaluate the trace and produce assessments

Trigger Evaluation

Workflow triggers check if they match the trace or its assessments

Node Execution

If a trigger matches, the workflow executes its nodes in sequence

Final State

Trace ends as Completed, Rejected, or waiting for human decision

Workflow Nodes

Build workflows using these node types:

Trigger Nodes

Triggers define when a workflow should start executing.

Policy Trigger

Fires when a specific policy matches. Configure which policy and what outcome (compliant/non-compliant).

Conditional Trigger

Fires based on custom conditions - check trace fields, arguments, or assessment data.
Policy Trigger Example:
Trigger when: GDPR Policy
Condition: Non-compliant
Risk Level: High or Critical
Conditional Trigger Example:
Field: trace.functionName
Operator: contains
Value: "refund"

Action Nodes

Action nodes do something when reached.

Human Intervention

Creates a review task and pauses the workflow until a human decides.

Notification

Sends an alert to configured channels without pausing.

Rejection

Marks the trace as rejected and ends the workflow.

Human Intervention Node

The human intervention node is where human review happens:
SettingDescription
Approvals NeededHow many people must approve
TimeoutHow long to wait before escalation
ChannelsWhere to send the approval request
Outputs:
  • Approved → Continue to next node
  • Declined → Execute decline path
  • Change Requested → Execute change request path
  • Error/Timeout → Execute error path

Notification Node

Send alerts without blocking:
SettingDescription
ChannelsSlack, Email, Teams, or SMS
Include DetailsWhether to include trace arguments
Include Policy ResultsWhether to show assessment data

Building a Workflow

  1. Go to Workflows in the dashboard
  2. Click Create Workflow or edit an existing one
  3. Drag nodes onto the canvas
  4. Connect nodes by dragging from outputs to inputs
  5. Configure each node by clicking on it
  6. Save and activate

Example: GDPR Review Workflow

[Policy Trigger: GDPR Non-Compliant]


[Human Intervention]
    │         │
    │ Approved│ Declined
    ▼         ▼
[Notification: Slack]  [Rejection]

Example: High-Value Transaction Workflow

[Conditional Trigger: amount > 10000]


[Human Intervention: 2 approvals needed]
    │         │
    │ Approved│ Declined
    ▼         ▼
[Notification: Email]  [Rejection]

Trigger Conditions

Conditional triggers can check:
CategoryFields
TracefunctionName, sourceName, direction, status
ArgumentsAny key in trace.args
MetadataAny key in trace.metadata
Assessmentconfidence, riskLevel, compliant, tags, recommendation
Operators:
  • equals, not_equals
  • contains, not_contains
  • greater_than, less_than
  • in, not_in

When No Triggers Match

If no workflow triggers match a trace:
  • Trace is automatically marked as Completed
  • No human review required
  • Assessment data is still logged

Best Practices

Start Simple

Begin with one trigger and one action. Add complexity as needed.

Use Policy Triggers

Let your policies do the heavy lifting. Trigger workflows based on policy results.

Set Timeouts

Always configure timeouts on human intervention nodes to prevent stuck workflows.

Test First

Send test traces to verify your workflow behaves as expected.