Beobachtbarkeit

Voice agent transcripts in Orchestrator and Maestro traces, plus AI Trust Layer governance and audit for voice.

Ablaufverfolgungen

Every voice call writes a transcript and a node-level execution record, whether the call was inbound or outbound. Open the same run in Orchestrator or Maestro.

Orchestrator

The conversation tree. Open Conversation > Exchange, select User message or Agent message > Content part, and read the transcript in Other attributes.

• The transcript sits on the conversation, not on the node.
• Exchanges are written when they end, not word by word, so the trace fills in a turn at a time.
• Interruptions are recorded. When a caller talks over the agent, the current exchange closes there, the transcript marks it [interrupted], and a new exchange opens.

If you want to get the conversation history for processing (such as semantic analysis) you can use the Get Conversation Context node. See Voice agent.

Maestro

View a Flow instance that was executed. Open it from the Flow's execution history.

AI Trust Layer

Voice calls reach the speech model through the LLM Gateway, so they pick up AI Trust Layer governance, guardrails, and audit the same way other UiPath products that call LLMs do.

Guardrails and governance

All AI Trust Layer policies apply to voice except for the Centralized guardrails tab. This includes masking personally identifiable information (PII) and content policies. The Conversational Agents toggles on that tab apply only to chat, not to voice.

However, UiPath enforces Zero Data Retention (ZDR) with all AI vendors, meaning no audio or transcripts are stored with our AI vendors such as OpenAI and Google.

Tool-level guardrails are available on the tools connected to the Voice agent node. See Tool guardrails.

Registerkarte	Voice
Product toggles	Applied
Zentralisierte Leitplanken	Not applied
Funktionssschalter	Applied
Modelle	Applied
Modellregionen	Applied

Audit

Real-time voice is audited as one record per conversation, not per turn. The conversation's full transcript sits on that parent record; turn-level detail is not shown in the audit view.

Feature toggles on the AI Trust Layer policy control whether inputs and outputs are stored. If you turn that off, the audit entry still appears but the payload is Blocked by Policy.

See viewing audit logs.

Zugehörig

• Observing runs — the execution trace in Maestro Flow.
• Über AI Trust Layer
• Einstellungen für AI Trust Layer-Richtlinien
• Tool-Leitplanken