End-user perceived latency metrics in shareable AI assistant demo links
17, Dec 2025
Shareable demo links for AI assistant widgets now display end-user perceived latency for each conversational turn. This exposes the full round-trip delay a user experiences, instead of platform-side timing that excludes last-mile effects.
What’s new
- Per-turn latency display: Shows latency after every user-agent exchange in shareable demo links.
- Client-side measurement: Latency is captured on the end user’s device, not Telnyx backend infrastructure.
- Full round-trip timing: Measures from when the user stops speaking to when agent audio begins.
- End-to-end scope: Includes turn detection, STT, LLM, TTS, tool calls, orchestration, and network transmission.
- Voice-relevant metric: Reflects the mouth-to-ear turn gap used to evaluate voice agent responsiveness.
Why it matters
- Reflects real user experience: Includes last-mile network latency that platform metrics and recordings miss.
- Prevents misleading comparisons: Backend or recording-based metrics systematically underreport perceived delay.
- Improves agent tuning: Helps isolate whether latency comes from models, tools, or network conditions.
Aligns with voice standards: Uses the most meaningful latency metric for conversational voice agents.
Example use cases
- Sharing accurate latency data during customer demos.
- Comparing perceived responsiveness across regions or devices.
- Explaining why live calls feel slower than platform recordings.
Getting started
- Log in to the Mission Control Portal
- Navigate to your AI assistant widget.
- Open or generate a publicly shareable demo link.
- Run a voice conversation and review per-turn latency shown in the demo UI.
Learn more by exploring AI assistant widgets in the Mission Control Portalor by contacting your Telnyx team.