Introduction
How the Parlot dashboard helps you observe production voice AI agents and nested text workflows.
The Parlot App Guide covers the web dashboard: fleet health, agent workspaces, session debugging, evaluations, exports, and organization settings.
Instrument voice AI agents (LiveKit, Pipecat) or LangGraph workflows with the SDK Quick Start, then use this guide to work with the data in the product.
What the dashboard is for
Parlot answers questions beyond the transcript alone:
- Is the agent completing the caller’s goal?
- Where did latency or cost spike in the turn pipeline?
- Which agent in a multi-agent flow caused a failure?
- What topics and outcomes are trending across the fleet?
Conversations and sessions
Parlot uses a two-level model:
- A conversation is the ongoing thread with a user on a channel (voice, webchat, text messaging, WhatsApp).
- A session is a bounded episode inside that conversation—the unit of analysis, evaluation, and cost.
Voice calls usually map one conversation to one session. Messaging channels can produce several sessions over time when idle timeouts close an episode. Configure those rules under Conversation boundaries.
Dashboard tour
| Area | What you do there |
|---|---|
| Overview | Org-wide quality, speed, and cost |
| Agents | Per-deployment health, topology, topics, and capture settings |
| Topics | Org-wide topic groups and drill-down into sessions |
| Sessions | Search, paste vendor IDs, open the session debugger |
| Cost & margin | Rate card, unpriced models, and margin by agent or client |
| Exports | Download session artifacts |
| Evals | Custom large language model (LLM) judges and run history |
| Settings | Members, plan & billing, recording, logs, privacy, API keys, and more |