The model is already on your Mac.
ParseMeet is built on Apple's Foundation Models framework — the language models behind Apple Intelligence, on your device and on Private Cloud Compute. That one architectural choice explains almost everything about how the app feels.
What it is
Apple Intelligence, as a foundation.
With macOS and iOS, Apple ships a large language model as part of Apple Intelligence and gives apps direct access through the Foundation Models framework. By default the model runs on the Neural Engine of Apple silicon — no API keys, no per-token billing, no AI vendor in the loop.
ParseMeet uses it for the hard part of meeting notes: reading a raw transcript and deciding what was a decision, what was a commitment, who made it, and when it's due. The framework's guided generation lets us define the exact shape of an outcome doc — sections, checklists, owners, dates — and have the model fill that structure directly. That's why ParseMeet docs come out structured instead of as a summary blob.
The bigger model
When a meeting needs more, it stays just as private.
An on-device model has a context window; a two-hour cross-functional review can exceed it. When a meeting calls for it, ParseMeet sends the transcript — never the audio — to the larger Apple model on Private Cloud Compute: Apple silicon servers, hardened OS, the same security architecture as the device in your hand. Per Apple, data sent to PCC is never stored, is used only to fulfill your request, and "isn't accessible to anyone other than the user — not even to Apple." It isn't accessible to us either — there's no ParseMeet server in the path, no API key, nothing for us to hold.
And it's verifiable: Apple publishes every production PCC build for independent security research, and your device refuses to talk to a server whose software hasn't been publicly logged. Apple sees only content-free request metadata — approximate size, feature, duration — never the content. Offline? Everything runs on-device, automatically. The full data map →
Why it matters
What Apple's private stack buys you.
Instant, not "processing…"
Cloud-vendor tools upload your audio, wait in a queue, and email you later. ParseMeet parses moments after you hit stop — the doc is ready when you hang up.
Free to run, forever
Vendor AI has a per-minute cost someone has to pay — usually you, as a subscription. Apple's models cost nothing per meeting, so ParseMeet is one purchase instead of a meter.
Private by architecture
Your audio never leaves your device. Your transcript is readable by exactly one party: you. Not us, not Apple, not an AI vendor. More on privacy →
Maintained by Apple
The models improve with OS updates, tuned for Apple silicon end to end. Nothing to download, configure, or keep compatible.
The honest tradeoff
The right model, pointed at the right job.
The on-device model is smaller than the frontier models running in data centers. For open-ended essay writing, that matters. For turning a transcript into a template's sections, it doesn't — the task is extraction and structuring, and guided generation constrains the output to exactly the shape your template defines. When a meeting genuinely needs a bigger model, ParseMeet reaches for the one cloud that can't read your data — instead of the many that can.
Requirements: an Apple silicon (M-series) Mac on macOS 26.4 or later, or an iPhone 15 Pro or later on iOS 26+, with Apple Intelligence enabled. Private Cloud Compute additionally needs an internet connection — without one, ParseMeet runs on-device.
Sources: Apple's Private Cloud Compute security overview · Apple Intelligence & Privacy · iCloud data security · CloudKit private databases
AI notes without the AI bill.
No metering, no per-seat AI surcharge — the models come with the platform. In beta now; one payment at 1.0.
Email me for beta access