Benten listens to your Voice AI agents — measuring latency, dead air, interruptions and acoustic quality across ElevenLabs, Vapi and Retell in one open-source dashboard.
Self-host in minutes · MIT licensed · No credit card
Health Score
+3.2 vs 7d
Turn Latency
-84ms vs 7d
Dead Air
-0.6% vs 7d
Barge-ins
+0.3% vs 7d
Live turn analysis
support-agent-v4 · call #a91f
Agent health
onboarding-bot
ElevenLabs
sales-outbound
Retell
support-agent-v4
Vapi
renewal-followup
Retell
triage-en
Vapi
Vendor-agnostic. Works with the voice stack you already use.
The problem
Traditional AI monitoring evaluates strings. Real callers judge you on timing, tone, and pauses.
A voice bot can generate a perfect response — but if it takes 3 seconds to reply, the caller hangs up thinking the line dropped.
Bots that interrupt mid-sentence, freeze in dead air, or speak over users quietly destroy brand reputation.
ElevenLabs, Vapi, Retell — every vendor gives you a different view. Nobody gives you the whole picture of voice quality.
Platform
One control room for every voice agent your team ships — no matter which provider it runs on.
Drop in API keys for ElevenLabs, Vapi or Retell. Benten auto-syncs every agent and assistant profile into one dashboard — zero manual setup per agent.
Turn latency, dead air %, barge-ins and a 0–100 health score derived from raw call audio — not just LLM logs.
Automatic alerts on latency spikes, barge-in bursts and audio drops — with root-cause drilldown across TTS, webhooks and LLM.
Historical trends, response-speed distributions, call volume, health scores and raw config — for every agent, in one click.
Workspace isolation, role-based access and audit-ready logs — built for agencies managing many client portfolios.
MIT-licensed. Self-host it, extend it, audit it. Own your voice observability stack end-to-end.
The metrics that matter
Speed
Exact millisecond delay from when the human stops speaking to when the agent responds.
Silence
Unintended silences during processing, hand-offs, or model warm-up.
Flow
Calls where the user had to talk over the agent, or the agent failed to yield the turn.
Overall
Response speed, conversational flow and acoustic quality synthesized into one benchmark.
How it works
Paste API keys for ElevenLabs, Vapi or Retell. Encrypted, workspace-scoped, revocable.
Benten catalogs every deployed assistant across every provider into a single unified list.
Raw call audio is analyzed for turn timing, dead air, barge-ins and speech dynamics.
Drill down into any call. Trace latency to TTS, webhooks or LLM — and fix it.
Track health score trends over time. Ship voice changes with confidence.
Built for
Measure real-world latency, fine-tune voice settings, and debug call bottlenecks across vendors.
Audit voice call quality at scale — without listening to thousands of hours of raw audio.
Track SLA compliance, reduce caller drop-offs, and safeguard CSAT.
Report on agent performance for multiple client organizations from a single dashboard.
Benten is fully open source. Self-host it on your own infra, extend the evaluators, contribute back — and never lock your voice quality data inside another vendor.
$ git clone github.com/benten/benten $ cd benten && docker compose up → syncing agents from ElevenLabs... → syncing agents from Vapi... → syncing agents from Retell... ✓ 34 agents discovered ✓ dashboard live at localhost:3000
Plug in your provider keys, and get a full acoustic QA report on every agent — today.