CAPTCHA for the web - Vocos Bouncer for AI voice agentsLearn more

The trust layer
for voice agents.

The trust layer for voice agents.

VERIFIEDINSPECTBLOCKED
VERIFIED0INSPECT0BLOCKED0

One challenge. Every call.
Maximum protection.

Vocos Bouncer screens every incoming call against dozens of signals - acoustic forensics, prosody, formants, temporal artifacts, and behavioral patterns. Humans pass. Synthetic voices don't.

<1s
avg latency

Sub-second bouncer

Every call screened in under a second. No friction for real callers - deepfakes blocked at the door, before they ever reach your agent.

/v1/detect
verdict

Drop-in bouncer

One endpoint. Send audio, get verified or blocked. Wire the bouncer into any voice agent in hours, not months.

Works with

VAPI
Retell
LiveKit
Twilio
& more

Sits in front of any voice agent

Vocos Bouncer drops into any voice stack. A few lines of code and every incoming call is screened before it hits your agent.

97%conf.
spectrogram analysis

Dozens of signals

Forensics 0.3B plus prosody, formants, spectrogram artifacts, and behavioral patterns. Confidence scores and explanations for every challenge.

Age 28MaleIntent: PurchaseEmotion: Calm

Beyond pass / block

Every cleared call comes with structured intelligence - age, gender, intent, emotion. Your agent doesn't just trust the caller, it knows them.

AgeGenderIntentEmotionVerification
Live threat landscape

Voice fraud is exploding.
Your agent needs a bouncer.

3 sec

to clone any voice with today's AI tools

1,300%

surge in deepfake fraud at contact centers in 2024

$12.5B

lost to contact center fraud in 2024

70%

of people can't tell a real voice from AI-cloned

One call. The bouncer is on the door.

From zero to fully protected in under five minutes.

cURLPython Beta
# One API call. Full voice security.
curl -X POST https://api.vocos.io/v1/detect \
  -H "X-API-Key: your-api-key" \
  -F "file=@audio.wav"

# That's it. You're protected.
{
  "verdict":     "real",
  "confidence":  0.973,
  "model":       "forensics-0.3B"
}
REST APIWebSocket50+ LanguagesForensics 0.3B>99% Accuracy

How it works

Incoming call → turnstile → your agent. Three steps.

01

Incoming call

Every call routed to your voice agent passes through Vocos Bouncer first.

02

Screen

Like a bouncer at the door: Forensics 0.3B plus dozens of acoustic and behavioral signals decide who's real.

03

Walk in or stay out

Real callers reach your agent untouched. Synthetic voices stay outside the door.

< 1s

Response time

0.0*%

Tested

Detection accuracy

Effortless

Integration

0 free

Free credits

Frequently asked questions

Everything you need to know about Vocos.

Today's AI can clone any voice in just 3 seconds. That's incredible technology - but it also means anyone can impersonate anyone. Voice agents are the new front door for businesses, and they need protection. Vocos exists to make sure your voice AI can't be weaponized against you.

Vocos is the only comprehensive solution for voice agent protection. We're purpose-built for voice agent security - not adapted from a generic audio tool. We detect the most recent TTS and voice cloning methods that others miss, integrate in minutes with any voice stack, and go beyond security with audio intelligence features like age, gender, and intent detection - so you know your caller and can adjust your flow in real time.

One credit equals one screened call. Pro costs $0.01 per credit. The free plan includes 100 free credits.

Under 1 second. Forensics 0.3B runs on dedicated infrastructure and returns a verdict with a confidence score in real time - fast enough to use inline during a live call, not just after the fact. We also support faster inference for special cases - contact us for more details.

Forensics 0.3B is trained on over 5 million audio samples spanning more than 100 of the latest TTS systems and attack vectors, including the voice-cloning models active in fraud attacks today. It excels on every public deepfake benchmark we've tested, achieving >99% on the In-the-Wild dataset (the industry-standard real-world test) and 99.5% on our internal 2026 evaluation set. That said, no detection model is perfect - Forensics can make mistakes, so treat its verdict as a strong signal alongside your other defenses, not the sole arbiter. The model is updated continuously to handle new threats, so individual benchmark scores can shift up or down over time.

Security is in our DNA. All communication is encrypted in transit, and your data is protected by enterprise-grade infrastructure.

We're deep learning researchers with elite military intelligence experience. We've spent years building AI systems in high-stakes environments, and now we're making that level of security accessible to every developer and business. Reach out: eliya@vocos.io.

Put a bouncer on your voice agent.

Start with our free playground. No credit card required. Go instantly from signup to fully protected.