Bring your own model
Summaries
on your terms.
Local Ollama for full privacy. Your own OpenAI key for sharper output. Never our cloud - Mac Note Taker doesn't run inference. You point it at the model you trust.
- Launch date locked - Tuesday next week, marketing site ships Monday night.
- Founder coupon - FOUDNER → $79 added to hero + pricing card.
- Counter pulls live from Stripe - max_redemptions = 100 (action: confirm).
- OG image - Lina to redo with new palette before Monday.
- Test purchase - Owen to run a real $0.79 charge end-to-end.
Action items
- ·Lina redo OG image · Mon
- ·Marko verify Stripe coupon cap · today
- ·Owen run end-to-end paid checkout · today
Pick your provider.
One dropdown in Settings → AI Assistant. Same prompt, same output schema, different backend.
Ollama (local)
Zero data leaves your Mac
http://localhost:11434/v1
Model
qwen2.5:7b-instruct
Best for
- NDA, HR, customer-sensitive meetings
- Offline / on a flight
- Cost-sensitive (zero per-meeting cost)
A 7B model on M3 Pro takes 8-15s to summarize a 30-min transcript. Output is 8/10 vs OpenAI's 9.5/10.
OpenAI
Your own API key, your own bill
https://api.openai.com/v1
Model
gpt-4o-mini
Best for
- Long, dense, technical meetings
- Sharpest action-item phrasing
- You already have an OpenAI key set up
Transcript leaves your Mac when summarization runs. Mac Note Taker doesn't proxy it - direct call from your Mac to the endpoint you set.
What the summary actually contains
5-bullet summary
What was decided, what changed, what's open. No fluff.
Action items
Bulleted by owner. The transcript is the source of truth; the action items are the audit trail.
Speaker rename
Speaker 1 / Speaker 2 → Anya / Marko / Priya, pulled from transcript context + calendar attendees.
Topics
3-5 topical sections so a 60-min call gets re-readable in 60 seconds.
Three privacy postures.
Pick the one your meeting needs. Switch per-meeting.
Strict - Ollama only
Audio, transcript, and summary all stay on the Mac. Recommended for HR, legal, NDA, GDPR-sensitive workstreams.
Standard - your OpenAI key
Transcript + prompt go to your OpenAI tenant. Audio still never leaves the Mac. Fine for most internal meetings.
Custom - any OpenAI-compatible
Groq, Together, Fireworks, Azure OpenAI, a self-hosted gateway. Same client; you provide base URL + key.
A summary in every meeting.
$149 lifetime · 3 Macs · code FOUDNER for $79.