What did this call cost?
Provider bills aren’t session truth. Need STT / LLM / TTS dollar breakdown and margin against what you charge — not a DIY rate card spreadsheet.
The gap
What production voice teams can’t answer today
Cloud Insights and generic LLM traces leave the questions that matter for voice unanswered.
Is the agent working—or just ending?
Completion rate isn’t an outcome. Score goals and call results, then separate agent stalls from audio degradation and latency.
Where does the data live after 30 days?
Your dashboard, exports, and shareable session review — not a Cloud retention cliff or another Postgres you have to wire yourself.
What Parlot gives you
Cost, quality, topology, and your archive
The surface voice-agent teams build by hand — already modeled for voice turns, not flat traces.
Cost
Dollar cost per session — STT, LLM, TTS
Bring your rate card and overrides. See stage costs on every call, then Cost & margin across agents and clients — so you stop reconciling UsageCollector against provider invoices alone.
Quality & Speed
Voice-native Quality and Speed
Outcomes and goals stay separate. Latency stacks, attention ranking, and turn timelines that understand interruptions and handoffs.
Topology
See which agent broke the call
Orchestrator, sub-agents, tools, and handoff order on one graph — with error markers on the edge that failed. Multi-agent LiveKit and Pipecat sessions, not a flat trace dump.
Your data
Your portal — past the 30-day Insights window
Sessions with summaries, async export, and share links. Works with LiveKit Cloud or self-hosted media — ship analytics to a place you own instead of rebuilding Agent Insights.
Ready to know whether the next call actually worked?
Request early access — we are onboarding waitlist invites first.
OpenTelemetry SDK for LiveKit, Pipecat, and LangGraph —
parlotize() once.