Skip to main content

Overview

Every AI interaction captured by Velatir follows a defined lifecycle. Understanding this lifecycle helps you interpret trace statuses and configure your agents effectively.

Lifecycle Stages

Received

A data source sends an AI interaction to Velatir. The trace is stored and marked as Received.At this stage, the trace data is captured and the platform assigns it to a session if related traces already exist. Processing begins immediately.

Agent Assessment

All active agents evaluate the trace in parallel. Each agent specialises in a different compliance domain and independently produces an intent.
AgentDomain
Data ProtectionPII, financial data, credentials, health information
GatekeeperAccess control, usage rules
Each agent returns one of three intents: Allow, Block, or Escalate. The trace moves to Assessed status once all agents have completed their evaluation.

Action Resolution

The platform resolves the combined agent intents into a single action based on each agent’s role.
RoleWhat the Agent Can Do
ObserverReports findings only. Its intent is logged but does not affect the trace outcome.
EnforcerCan block traces or require approval. A Block intent rejects the trace. An Escalate intent requires human approval.
The most restrictive applicable action wins. If any Enforcer blocks, the trace is rejected. If any Enforcer escalates, a review task is created.

Final State

The trace reaches one of three final states based on the resolution.
StatusWhen It Happens
CompletedAll agents allowed the trace, or a human reviewer approved it
RejectedAn Enforcer blocked the trace, or a human reviewer rejected it
EscalatedA review task has been created and the trace is waiting for a human decision
Once a human resolves an escalated trace, it moves to Completed or Rejected.

Trace Directions

Every trace has a direction that describes the type of AI interaction it represents.
DirectionDescriptionExample
InletAn incoming request to an AI systemA user submits a prompt to a chatbot
ResponseAn outgoing response from an AI systemThe chatbot’s reply
SignalA passive eventA tool call, system log, or background process

Sessions

Related traces are automatically grouped into sessions. A session represents a complete conversation or workflow. For example, a user’s full chat exchange with an AI assistant. Sessions give reviewers the full context when evaluating an escalated trace. Instead of seeing a single interaction on its own, they can view the entire conversation.

Monitoring Traces

Track trace progress in the web dashboard.
  • Traces view shows the status and agent assessments for all traces in a workspace
  • Trace detail lets you inspect a specific trace to see each agent’s evaluation, the resolution logic, and the final outcome
  • Sessions view shows all traces within a session for complete conversation context
  • Review Tasks lets you manage traces that are waiting for human decisions

Sessions

Understand how traces are grouped into sessions.

Review Tasks

Understand how escalated traces are reviewed and resolved.