Public beta · v1.0.2

Records everything.
Keeps nothing.

Always-on macOS meeting recorder with on-device voice fingerprinting. Audio deletes every night. Only structured data survives.

⚠ Public beta. Expect rough edges. Free while we shake them out.

macOS 15+ Apple Silicon Notarized Free during beta
What's different

Six architectural bets.

Not incremental improvements. Different assumptions.

01Capture

Always-on

Runs silently from login. Scheduled calls, Slack huddles, the 3-minute chat where a decision gets made — all captured automatically.

02Speakers

VoicePrint — persistent identity

256-dim WeSpeaker embeddings. Confirm a voice once, FiftyX recognizes them in every future conversation — not just this session.

03Retention

Ephemeral audio

Audio auto-deletes at end of day. Transcripts, summaries, and embeddings persist. Privacy by deletion, not encryption.

04Context

Calendar-aware

Nylas reads Google and Microsoft calendars. FiftyX knows who's expected and matches voices to attendees before you tag anyone.

05Hardware

No hardware needed

Uses your Mac's built-in mic — or AirPods, headset, whatever you already own. No pendant to charge. Nothing to forget.

06Models

Bring your own model

Claude, OpenAI, Gemini, or Grok — your key, your choice of cost and quality. Text-only payloads; audio never leaves your Mac.

Inside FiftyX

Every step runs on your Mac.

Capture → VAD → transcription → diarization → summarization. Only the last step is optionally cloud, and only with text — never audio.

Capture
Core Audio Tap + AVAudioEngine
sys audio + mic
16 kHz mono
Gate
Silero VAD
skip ~65% silence
GPU-idle
Transcribe
WhisperKit · small
Core ML + Metal
30s chunks
Diarize
pyannote + WeSpeaker
256-dim embeddings
on Neural Engine
Summarize
Your LLM · your key
Claude · OpenAI
Gemini · Grok
On-device Optional · your key Audio deleted at end of day
The science behind VoicePrint
11.2% DIARIZATION ERROR RATE

Best independently verified accuracy in the industry. Built on pyannote.audio — the most-cited speaker diarization framework in academic literature, from CNRS and peer-reviewed at Interspeech and ICASSP. Runs locally through Apple's Neural Engine.

CNRS · research lab Interspeech · ICASSP WeSpeaker · 256-dim Neural Engine
Not the same category

How FiftyX compares.

Dimension FiftyX Granola Otter Gong Plaud
Always-on recording Yes No No Sales only Hardware required
Speaker recognition Persistent, on-device Per-session Per-session Per-session Per-session
Audio retention Deleted daily Stored Cloud Cloud On device
Platform Native macOS Cross-platform Web / mobile Web Hardware + app
Audience Everyone Everyone Everyone Sales teams Everyone
Price Free (beta) Free · $12/mo $17–40/mo Enterprise $159+ · $79/yr
LLM Bring your own Built-in Built-in Built-in Built-in
Bring your own model

Your key. Your cost. Your choice of quality.

FiftyX calls the model provider directly — no backend proxy, no hidden middleware, no per-seat LLM markup. Swap providers anytime in Settings.

Only transcript text leaves your Mac. Audio never does.

Claude · Opus 4.6 OpenAI · GPT-4.1 Gemini · 2.5 Pro Grok · 4
Privacy by deletion

We don't encrypt more. We keep less.

Deleted every day

Audio files auto-delete at end of day. Only transcripts and summaries persist — plain text, fully searchable, local to your Mac.

  • Audio deleted at 00:00 (configurable)
  • Pause anytime from the menu bar
  • Retroactive deletion — "delete the last 5 min"
  • App exclusions — Signal, WhatsApp, etc.

Audio never leaves your Mac

Transcription happens on-device via WhisperKit. When you use an LLM for summaries, only transcript text is sent — never audio.

  • On-device transcription · WhisperKit
  • On-device speaker diarization · pyannote + WeSpeaker
  • Your API key, your choice of LLM provider
  • Transparency dashboard shows every byte stored
Public beta · v1.0.2

Try it.

macOS 15 Sequoia or later, Apple Silicon. Notarized DMG, direct download.

Download for Mac

This is a public beta. The pipeline is solid — capture, transcription, voice ID, knowledge graph all working — but the polish isn't finished and you'll find bugs. Auto-updates ship via Sparkle so fixes land within hours of being shipped. Free while we're in beta. Email hello@50x.ing when you find something broken.