Real-time call intelligence

Stop scams.
Live.

Phigil listens to live voice calls and flags scams, fraud and threats as they are spoken — in roughly 80 milliseconds, on your own infrastructure. The same platform handles age assurance and identity verification.

Book a demo See how it works

Built for telcos, marketplaces & platforms at national scale.

$0T lost to scams worldwide in 2024.
Most of it started with a conversation.

Source: Global Anti-Scam Alliance, Global State of Scams 2024 — est. US$1.03 trillion.

How it works

Three steps, on every call, in real time.

  1. 01

    Listen

    Stream call audio into Phigil over your existing stack — SIP, WebRTC, or our SDK. No call leaves your network unless you want it to.

  2. 02

    Understand

    Each utterance is transcribed and classified for scam patterns, social-engineering and threats — roughly 80 ms per segment, CPU-only.

  3. 03

    Act

    Warn the caller, alert an agent, or block the line — with a category, confidence score and a plain-language explanation for every flag.

Capabilities

One trust layer. Several signals.

Flagship 01

Real-time scam & threat detection

Catch impersonation, authorised-push-payment fraud and coercion while the call is still happening. Per-speaker transcripts, category, confidence and an explanation — streamed live, not in a nightly batch.

  • ~80 ms per utterance
  • Per-speaker attribution
  • Explainable verdicts
02

Age assurance

Estimate age from a single image or a live camera frame — for onboarding, age-gating and compliance, without storing documents.

03

Identity verification

Match a face against an enrolled reference to confirm a person is who they claim to be — at sign-up or step-up.

04

Developer platform

One authenticated API, scoped keys with rate limits, and full tracing and metrics. Ship a proof of concept this week.

Built for scale

Engineered for the people who can't afford a false sense of security.

Runs in your VPC

Deploy on-prem or in your cloud. Voice data stays inside your perimeter — a requirement, not an upsell.

No GPU fleet required

Quantised ONNX models run CPU-only, so cost scales with traffic instead of with a rack of accelerators.

Observable by default

OpenTelemetry traces and Prometheus metrics out of the box. You see every decision and every millisecond.

Privacy by design

Process in-stream, retain nothing by default. Tune retention to your jurisdiction and your policy.

0ms per-utterance verdict
0% in your perimeter
0 signals, one API
CPU no accelerators needed

FAQ

Questions buyers actually ask.

How does Phigil integrate with our voice stack?

Over the interfaces you already run — SIP/RTP, WebRTC or our SDK. Phigil sits in the media path or taps a fork of it, and returns verdicts on a stream you consume from your IVR, agent desktop or fraud engine.

Does our customers' audio leave our network?

Only if you choose to route it that way. Phigil is designed to run inside your VPC or data centre, so voice never crosses your boundary by default.

Is it fast enough for a live call?

Yes — classification runs in roughly 80 ms per utterance on CPU, which is well within the window to warn a caller or agent mid-conversation.

How do you handle false positives?

Every flag carries a category, a confidence score and a plain-language explanation, and the decision threshold is yours to set — so you tune the trade-off between catching scams and interrupting good calls.

What else can the platform do?

The same authenticated API offers age assurance and identity verification, so one integration covers onboarding, age-gating and live-call defence.

Bring scam intelligence to every call.

Tell us about your traffic and we'll stand up a proof of concept on your stack.