Skip to main content

​
Overview

When a trace enters Velatir, it goes through several stages before reaching its final state. Understanding this lifecycle helps you configure policies and workflows effectively.

​
Lifecycle Stages

Trace Created → Policy Assessment → Workflow Execution → Final State

​
1. Trace Created (Status: Received)

Your system sends a trace to Velatir through the API, SDK, or integration. The trace is stored and marked as Received. At this point:
  • Trace data is captured
  • Session grouping happens
  • Processing begins

​
2. Policy Assessment (Status: Assessed)

Active policies evaluate the trace. Each policy produces an assessment with:
FieldDescription
CompliantDoes the trace comply with the policy?
Risk LevelLow, Medium, High, or Critical
ConfidenceHow confident is the assessment (0-1)
ReasonExplanation of the assessment
RecommendationSuggested action
TagsLabels for categorization
If no workflow is configured, the trace stays at Assessed status. You can review assessments manually in the dashboard.

​
3. Workflow Execution

If a workflow is active, triggers are evaluated against the trace and its assessments. Trigger Types:
TriggerEvaluates
Policy TriggerChecks if a specific policy matched
Conditional TriggerChecks trace fields, args, or assessment data
If no triggers match: The trace is auto-approved and marked Completed. If a trigger matches: The workflow executes its nodes sequentially.

​
4. Node Execution

Workflows contain nodes that process the trace:
Node TypeWhat It Does
Human InterventionCreates a review task, pauses workflow
NotificationSends alert to configured channels
RejectionMarks trace as rejected

​
5. Human Intervention (If Required)

When a workflow hits a Human Intervention node:
  1. A Review Task is created
  2. Notifications are sent to configured channels (Slack, Email, Teams)
  3. Workflow pauses until a decision is made
  4. Reviewer approves, declines, or requests changes
  5. Workflow continues based on the decision

​
6. Final State

StatusMeaning
CompletedTrace approved (auto or by human)
RejectedTrace rejected by workflow or human

​
Example Flow

1. User asks chatbot about competitor pricing
   └─ Trace created (Inlet, Received)

2. GDPR policy evaluates
   └─ Assessment: Compliant=false, Risk=High, Reason="Potential competitive intelligence"

3. Workflow trigger matches (GDPR High Risk)
   └─ Human Intervention node activated

4. Review task created, Slack notification sent
   └─ Manager sees: "High-risk trace needs review"

5. Manager approves with note
   └─ Decision recorded: Approved

6. Trace marked Completed
   └─ Audit trail complete

​
Processing Modes

​
Synchronous (Workflows Active)

When a workflow is configured:
  • Policies evaluate immediately
  • Workflow executes in-line
  • API response includes status and review task ID (if created)
{
  "traceId": "trace-123",
  "status": "RequiresIntervention",
  "reviewTaskId": "task-456"
}

​
Asynchronous (No Workflows)

When no workflow is configured:
  • Trace is queued for assessment
  • API response returns immediately
  • Check trace status later via API or dashboard
{
  "traceId": "trace-123",
  "status": "Received"
}

​
Monitoring Lifecycle

Track trace progress in the dashboard:
  1. Traces view - See status of all traces
  2. Trace detail - View assessments and workflow execution
  3. Sessions view - See all traces in a conversation
  4. Review Tasks - Manage pending human decisions

Understanding Traces

Learn about trace properties

Configure Workflows

Set up workflow triggers and nodes