Overview

The voice observability feature allows you to observe all the conversations that your agent does. You can treat it just like any other observe project, run evals and set up alerts for the same

What it is

Voice observability is the Observe feature that ingests call logs from supported voice providers (Vapi, Retell, ElevenLabs) and turns them into conversation traces in an Observe project. You do not instrument your voice app with the Future AGI SDK—you connect a provider using the provider’s API key and assistant ID (e.g. when creating or updating an agent definition). Call logs are then pulled in via polling or webhooks, normalized, and stored as traces with a root conversation span so they appear in the same Observe UI as other traces. You can run evaluations, set alerts and monitors, filter, and export. Each call typically includes transcript, recording URLs (e.g. mono/stereo), cost, and duration so you can debug and monitor voice agents in one place.

Purpose

  • Voice call visibility — See all conversations for a voice agent in one Observe project without adding SDK instrumentation.
  • Evals on voice — Run the same evals (e.g. quality, bias, adherence) on conversation spans from voice calls.
  • Alerts on voice — Use monitors and alerts on voice-project metrics and evals so you get notified when something degrades.
  • Transcripts and recordings — Access transcript and recording URLs (e.g. customer/assistant or stereo) from the trace or call log view for debugging and review.
  • Multi-provider — Support for Vapi, Retell, and ElevenLabs so you can observe agents regardless of provider.

Getting started

Set Up Voice Observability

Step-by-step setup: create or edit an agent definition, enable observability, and provide the provider API key and assistant ID. A project is created (or linked) in Observe and call logs are ingested. Includes supported providers.
Was this page helpful?

Questions & Discussion