Hapi on a 16GB Mac: What Works Fully Local (and What Doesn't)

If you've got a 16GB Mac, you've probably wondered whether Hapi's local AI features actually work on it, or whether you're stuck with a stripped-down version. Short answer: almost everything works, fully offline. This guide lays out exactly what to expect — no tech jargon, just what you can and can't do — and what your other option (Hybrid mode) actually means for your privacy.

The quick answer

	16GB Mac, fully local
Record and transcribe meetings	✅ Works
Meeting summaries & action items	✅ Works
Speaker identification	✅ Works
Chat with your meetings, mail, and calendar	✅ Works
Prep briefings before a meeting	✅ Works
Screen & Memory Assistant — simple requests	✅ Works
Voice notes	✅ Works
Background processing of older recordings	🟡 Works, one at a time (a bit slower)
Screen & Memory Assistant — multi-step requests	🟡 Works, less reliably
Voice notes — automatic title	❌ Skipped
Voice notes — automatic person/topic tagging	❌ Skipped
Live web lookups in chat ("what's today's score?")	❌ Not available

If your Mac has 24GB or more, every row above is a plain ✅ — the rest of this guide explains why 16GB draws the line where it does, and what to do about the ❌ rows if they matter to you.

What works exactly the same as on a bigger Mac

Meetings are the core of Hapi, and they get full treatment regardless of RAM: recording, transcription, who-said-what, the meeting summary, action items, the auto-generated title, and chat where you can ask questions about what was discussed, search your mail, or check your calendar. Prep briefings before a meeting also work fully local. None of this is watered down on 16GB — it just runs on a smaller on-device model than a 32GB Mac would use, which in practice you won't notice for everyday meetings.

What's a little slower, not broken

Two things change pace on 16GB, neither of which removes a feature:

• Catching up on old recordings happens one at a time instead of in batches, so a big backlog clears more gradually.
• The Screen & Memory Assistant (the quick-answer popup) handles simple requests fine, but can stumble on requests that bundle two things together — "summarize this and also draft a reply." Ask it one thing at a time and it holds up well.

What's turned off on 16GB, and why

Two voice-note extras are skipped specifically to protect your Mac while Hapi is thinking: automatic titles for voice notes, and automatic tagging of the people or topics mentioned in them. Your voice note itself is recorded, transcribed, and searchable exactly as normal — it just keeps its default name and doesn't get auto-tagged. (Meeting recordings are unaffected — this only applies to quick voice notes.)

Live web lookups — asking Hapi something that needs a real-time answer from the internet — aren't available fully local at all. That one genuinely needs a connection to work, on any Mac.

Why does 16GB draw a line here?

macOS is protective of memory. If Hapi's on-device AI, your browser, Slack, and Zoom are all fighting for the same 16GB at once, the operating system will forcibly close whichever app is using the most memory to protect the rest of your Mac — including, potentially, Hapi itself, mid-meeting. Skipping a couple of "nice to have" voice-note extras is a small trade Hapi makes so that the features that matter most — your meetings — stay reliable no matter what else is running.

Your other option: Hybrid mode

If you want those last few features, or you just want things to feel snappier, Hapi has a second mode: Hybrid. You keep recording and transcription 100% on your Mac — that never changes — but for the "thinking" steps (summaries, action items, chat answers), Hapi can hand the text off to a cloud AI model, using your own account with a service called OpenRouter.

This is entirely opt-in. Hapi ships local-first by default; Hybrid is something you turn on, not something that turns on for you.

Setting up Hybrid mode

1. Open Hapi → Settings → AI.
2. Switch the mode from Local to Hybrid.
3. Create a free account at openrouter.ai and generate an API key at openrouter.ai/keys.
4. Paste the key into the field in Hapi's AI settings.
5. Pick a quality preset if asked — the default is a sensible starting point.

That's it. From then on, the steps that benefit most from a bigger model run in the cloud; everything else — audio, transcription, your database — stays exactly where it was.

What Hybrid mode actually sends, and what it never does

This is the part worth reading carefully before you flip the switch.

Never leaves your Mac, in either mode:

• Your raw audio recordings
• Speech-to-text transcription itself
• Speaker/voice recognition
• Your entire Hapi database

Only in Hybrid mode, only the text you're actively processing:

• The transcript text of a meeting, when generating its summary or action items
• The question you type in chat, and the meeting/mail/calendar text needed to answer it

How your key and data are handled:

• Your OpenRouter API key is stored in the macOS Keychain — the same secure vault macOS uses for your other passwords — never as plain text, and never sent to Hapi's own servers. You're billed directly by OpenRouter, not by Hapi.
• Hapi defaults to zero-retention providers: models that contractually don't store or train on what you send them. This is on by default, not something you have to remember to enable.
• You can switch back to Local at any time in the same Settings screen — nothing about your past data changes when you do.

The short version: Hybrid mode trades "some meeting text, sometimes, goes to a provider you chose and pay directly" for a faster, more capable assistant. Your audio is the one thing that never leaves your Mac, in either mode.

Bottom line

On a 16GB Mac, Hapi's core job — recording, transcribing, and understanding your meetings — works exactly as well as it does on any other Mac, fully offline. The only things you give up are two small voice-note conveniences and live web lookups, and both are one Settings toggle away if you decide the trade-off is worth it to you.