In a technical sense it's absolutely possible. Owning the servers != transferring keys to the servers. Most E2E apps run both client and servers, it's about if they ever had key access.

That said, a pure c framework is probably a cleaner way to go.

I got tired of Electron meeting-transcription apps that upload my private audio to a server to do transcription/summary my MacBook could do locally in seconds. So I built Biscotti: a free, local meeting recorder. Calendar integration, voice isolation, speaker ID, AI summaries, native UI — and nothing ever leaves your device.

It's completely private: all AI runs locally, all data stored locally, nothing leaves your device, no cloud. It's native macOS (SwiftUI, not Electron), so it's smaller, faster, and more intuitive. And no bot ever joins your call - it uses system audio APIs, so it works with every app.

It integrates deeply with your Mac. It watches for meetings via audio APIs and offers to record when one starts, then automatically stops when the meeting ends (cancelable). It reads your calendar to see upcoming meetings, offers to start recording when they begin, and uses that metadata to help identify speakers. Speaker ID combines diarization, calendar metadata, and the transcript to figure out who's who automatically. Voice isolation separates your mic from your speaker output for clean audio.

The AI is customizable: want it to generate action items? A Slack-style summary? Talk like a pirate? Just edit the summary prompt. There's also markdown notes, and it keeps compressed audio files so you can re-transcribe later.

It's completely free. Under the PolyForm Perimeter License — fairly permissive (but not OSI open source).

Models and SDKs:

  Transcription — Whisper Large V3 Turbo via WhisperKit
  Summarization — Gemma 4 via llama.cpp (12B or E2B depending on your memory/processor)
  Speaker ID (diarization) — Pyannote via SpeakerKit

  Swift and SwiftUI
  XPC services for all AI processing, so the app stays tiny and stable and AI memory releases immediately
  System APIs: CoreAudio for audio, EventKit for calendars

Comments URL: https://news.ycombinator.com/item?id=48894351

Points: 1

# Comments: 1

> Don't turn a one-off or area-specific comment into a durable memory without my explicit confirmation. You have a history of over-indexing on one-offs, and those memories end up getting cited to override well-tuned skills.

A few people are running highly quantized models with limited context windows. It's still impressive, but not the benchmark level intelligence. Very few people could afford a rig for reasonable local performance at a reasonable quant, at full context size.

The antirez example is 2.6bit quant, 32k context, and few tokens per second... on a ~$7000 MacBook M5 (new RAM pricing).

Software is one domain where it excels because of structured training data and simulation environments, so I'm well aware it's better here than other areas.

Still there's somewhere balanced between saying every time it's "insufficient or incomplete or outright misleading" and "just trust AI". AI's a useful source of information/reasoning/research, but know you need to validate it's answers for important decisions.

If I take the first chapter of Harry Potter and write a new ending, it's still copyright violation. They took many parts.

Comments URL: https://news.ycombinator.com/item?id=48672088

Points: 2

# Comments: 0

I might build one for fun, but it's not going to change the economics alone. Still exciting it's possible.