Skip to content

Releases: johnny4young/portavoz

Portavoz 0.1.0 — knows who said what, locally

Choose a tag to compare

@johnny4young johnny4young released this 10 Jul 19:07

First public release. Portavoz records your meetings, transcribes them live, tells apart every voice — including yours — and summarizes everything 100% on your Mac. No cloud, no bots joining calls, no accounts.

Highlights

  • Dual-channel capture — your mic and the meeting audio, never mixed: your words are yours by hardware truth.
  • Live transcription under a second (Parakeet on the Neural Engine), with a one-click Whisper quality re-pass presented as a draft you approve.
  • Speaker separation on-device (DER 7.6% on an AMI sample) with live S1/S2 labels and optional voice enrollment so "Me" follows you to any microphone.
  • Summaries with three local engines: Apple Intelligence, Ollama, or the built-in Qwen3.5-4B (one 3 GB verified download, zero installs) — plus optional BYOK.
  • Meeting intelligence: typed summaries (standup/1:1/planning/interview), meeting health, cross-meeting to-dos, a local RAG chat over your history, and an MCP server for your AI tools.
  • Calendar-aware: pre-meeting briefs with cited "what to know", a heads-up banner, and recordings born with the real event name.
  • Live Companion (opt-in): cards that answer factual questions the room just asked, and a ping when someone says your name.
  • Audio first-class: synced lyrics-style player, channel-colored waveform, clips, AAC compression, drag-and-drop import.
  • English/Spanish UI, mixed-language meetings preserved per speaker.

Numbers (measured, reproducible with the dev CLI)

Stage Measured
Live finalization lag p50 0.07 s · p95 0.68 s
Live under batch load p95 0.53 s (target < 2 s)
Whisper refine 23–42× realtime
Diarization DER 7.6% (AMI sample)
Dual-channel drift 4 ms over 30 min
Cold start 0.94 s
Search across 1,000 meetings p50 22.8 ms

Install

brew tap johnny4young/tap
brew install --cask portavoz

Or download the notarized DMG below. First recording asks for microphone and "Screen & System Audio Recording" permissions — a guided first-run walks you through everything.

Requirements

macOS 14.4+ (Apple Silicon recommended; Apple Intelligence features need macOS 26).