Skip to main content

Overview

Scryon is a privacy-first call intelligence backend. You upload a phone-call recording and Scryon produces:

  1. A speaker-attributed transcript with stable speaker IDs, time-coded segments, and word-level confidence.
  2. A structured analysis of the call — a summary, key points, action items with owners and due dates, sentiment, topics, and follow-ups.
  3. Action items persisted as first-class objects you can list, complete, and snooze.

Everything runs as a single Spring Boot service backed by Postgres for state and S3-compatible object storage for artifacts.

Who Scryon is for​

  • Sales teams that want call summaries and action items without a meeting bot or screen recorder.
  • Field-ops and CX teams that record customer calls on Android and want them mined for insight.
  • Builders integrating call intelligence into their own products via a REST API.

Design principles​

  1. Privacy by construction. Raw audio is never persisted — it lives in-memory only long enough to be processed and then deleted from the temp-audio bucket within hours. See Privacy & security.
  2. Provider-agnostic pipeline. Diarization, transcription, and analysis each sit behind a small interface so a provider swap is a config change.
  3. Async by default. The HTTP layer returns a callId in milliseconds; the actual work happens on a background worker with idempotent state transitions.
  4. No hallucination of identity. Speakers are only named when the evidence (call metadata + transcript text + optional voice match) is strong. Weak evidence → Speaker N at LOW confidence.
  5. Observability is not optional. Every pipeline stage emits a structured event, a metric, and (when configured) an OpenTelemetry span. Sentry receives privacy-scrubbed errors.

End-to-end flow​

Each stage is documented under Features. The runtime orchestration lives in CallProcessingService.

What's next​