Long transcripts break the comfort zone of purely local inference. This clip explains how Apple routes heavy summary work through Private Cloud Compute while keeping capture and raw audio on your Mac.
Typical cloud AI retains copies on vendor infrastructure. PCC is built for stateless runs: the Mac encrypts transcript text, processing happens inside isolated enclaves sized for large context (including roughly 80,000-character meeting logs), and session artifacts are discarded after the response returns. CoreRecord for Mac uses that hybrid path for executive summaries over long sessions without treating a public cloud like shared storage.
Key Takeaways
- Secure courier, not a locker: PCC processes encrypted transcript payloads in isolation instead of hoarding user content on third-party servers.
- Large context on demand: Server-class enclaves handle summary prompts that exceed practical on-device context for full meeting transcripts.
- Local-first capture: Raw meeting audio stays on Mac hardware; the optional cloud step applies to sanitized text you explicitly request for summarization.
- CoreRecord workflow: Live transcription stays offline on device; AI summaries opt in through Apple’s PCC stack when you need depth over a long transcript.
- Series context: Part 1 covered AFM 3 on-device dictation; this clip is Part 2 on hybrid intelligence. The full PCC breakdown walks through routing, fallback, and privacy guarantees.
