The Otter quote landed, I did the math, and closed the tab.
Somewhere between the free tier's 300 transcription minutes and the Pro plan that quietly trimmed 6,000 minutes down to 1,200 (without dropping the price, by the way), I stopped caring how good the summaries were. AI meeting notes cost somewhere around $8 to $20 a seat per month at the tier that's actually useful, and that's just the subscription. This is roughly $100-$200 a year for one person's meeting transcripts, which is also about the point where I have to explain to the ops budget why a notes app is eating a line item.
And every one of those minutes talks to the cloud.
When you add up AI meeting notes cost, the per-minute caps matter
If you have real meetings, the pricing pages lie in a particular way. Otter Pro gives you a conversation cap and burns training minutes, Fireflies has a per-seat poof that doubles as you add people. The math stops working somewhere in the first busy month. My meetings are too long and too many to fit a 300-minute free tier and the minute caps are the part nobody budgets for.
I went to the other end: running it on the machine I already own.
What I pay for local notes now
Transcription is faster-whisper, running on my own CPU, free to the licensing gods. The minutes themselves come from a small model loaded in LM Studio - also free, also on my disk. I recorded a Teams call including the system side of the audio through WASAPI loopback, because Windows won't capture its own speakers unless you ask. No cloud account anywhere. No bot joins the call and nobody on the other end gets a notification that some vendor is eavesdropping for them.
The only real software I pay for is the wrapper that glues it into a usable Windows app, and that's a $9 one-time fee, not a subscription. After a year of clouds, one-time cost is the entire win.
The minutes come back structured - Summary, Key Decisions, Action Items with owners and dates where anyone said them out loud, Open Questions - and you can push them out as Markdown, plain text, or PDF. That covers a pretty long month of design reviews without a credit card attached.
If that's the direction you want to check out, it's at symshah.gumroad.com/l/xgoypx . It wants Windows 10-11 and Python 3.10+, LM Studio with any small model loaded. A GPU makes Whisper faster but a CPU finishes the job.
FAQ
How much do AI meeting notes tools cost?
Established cloud note takers run roughly $10-20 per user per month, so $120-240 per seat, per year, plus minute caps and bot recordings. A local setup is a one-time fee because the whole thing runs on your own machine.
Are AI meeting notes private if they're local?
If the app runs you record, transcribe, and summarize on your own PC with no account, then yes - Whisper and your local LLM never send the audio anywhere. Nobody can copy what never left the machine.
Do I need a GPU for local meeting notes?
No. Faster-whisper runs on CPU; a GPU just transcribes faster, so don't buy hardware for the notetaker. On older Intel/AMD chips, expect a longer wait after the meeting, but the output is the same.
Top comments (0)