← All ideas
EssaysNov 10, 2026Sam Gaddis

Foundation Calls: Meeting Notes With No Fee Per Person

Foundation Calls records calls on your own computer, writes the transcript and the summary, and files each call under the right customer in Foundation.

Granola's Business plan costs $14 a person a month as of November 2026. A 40-person company pays $6,720 a year for meeting notes. I was a happy customer, and I recorded more than 1,100 calls in it.

Film / Foundation Calls, 1:16

A speech model that takes up about 120 megabytes can now write the transcript on a laptop while the call is still going, and that costs nothing per call. The summary needs a language model. Each person can use the ChatGPT plan they already pay for, or Apple Intelligence, which comes on recent Macs.

When both of those cost little or nothing per call, the fee per person is paying for the app around them. So we built that app into Runpoint Foundation, the company operating system we give our clients with no license or seat fees. It's called Foundation Calls, and there is a 76-second film of it.

What it does

It listens from your computer, so nothing joins the meeting. The transcript appears as people talk. When the call ends it writes a summary of what was decided and the next steps.

It can also tell the voices on a group call apart and put a name on each one. The names come from the meeting app. In our one test, a Zoom call, it named all three remote speakers correctly. We have not tested Google Meet end to end yet.

The same lines of a call, twice. During the call the far end is one label, Speaker 2, and the text has no punctuation. A minute after it ends, the two remote voices are named Tom Okafor and Priya Shah and the text is punctuated. Sample call.
Fig. 01 / The same lines, before and after the voices are told apart

That part is not free. Turning it on sends each call's recording to a transcription service, AssemblyAI, which costs about 17 cents per hour of audio. At 20 hours of calls a month that is $3.40 a person. It passes Granola's $14 at about 82 hours.

A chart of dollars per person per month against hours of calls per person per month. Granola Business is flat at $14. Foundation Calls on the laptop is flat at $0. Foundation Calls with named speakers rises at 17 cents an hour and crosses $14 at 82 hours.
Fig. 02 / Paying by the hour beats a $14 seat until 82 hours of calls a month

Where the notes go

The filing matters more to me than the price. Foundation files each call under its customer, and you can move one it gets wrong. The person who recorded the call, the people assigned to that customer and company administrators can read the summary and the transcript. An assistant such as Claude can read them too, with the same permissions as the person asking. If someone asks what we promised a customer last week, they can look it up.

A call's page in Foundation: its title, date, length and who recorded it, a one-line summary, tags naming Foundation Calls and the customer's deal, the people on the call, and the decisions made. Sample call.
Fig. 03 / The call in Foundation, filed under its customer

It is for company calls. Administrators can read what it records, so it is the wrong tool for a private one.

What you can change

The transcript model and the summary model are two settings. When a better or cheaper one comes out, you change the setting and keep everything else.

The Transcription setting in Foundation Calls, with its list of models open. Three choices are marked On device, and AssemblyAI and Deepgram are listed below them.
Fig. 04 / Three ways to write the transcript on the computer itself

It also imports your own calls from a Granola export, so the old notes come with you. Imported notes stay on your computer, and only new calls go to Foundation.

What it doesn't do yet

It runs on Mac and on Windows PCs. On a Mac it needs an Apple chip, which means late 2020 or newer, and macOS 15 or later. Speaker names don't work in Teams. I don't know yet how the laptop model holds up when eight people talk over each other.

Foundation Calls is built on Anarlog, an open-source note-taker, which is why it took us under two weeks.