BOSWELL · THE ONE WHO KEEPS THE RECORD

The best meetings happen
in a room. Then they vanish.

Three people, a flip chart, a long meeting. Good ideas, a few wrong assumptions, and by morning nothing left but a photo of the whiteboard. Boswell sits on the table, keeps the record, and when someone asks it a question, it answers. Out loud, in the room, while the question still matters.

A name and a design · not yet built

James Boswell followed Samuel Johnson around London and wrote down what he said. Two centuries later, his name still means the one who keeps the record of the conversation.That is the whole job description

The question nobody asked

Someone at the table: Six of these boards, each running its own specialist model.

You: Boswell, what can one of them actually run?

Boswell: Eight gigabytes, shared with the operating system. Realistically a three-to-four-billion-parameter model per board. And the six boards can't pool their memory.

The answer is real; the exchange is reconstructed. In a real meeting this fall, that answer arrived the next morning, after the plan had been sketched around the wrong assumption. Two other premises fell the same way. Each one was a question someone could have asked out loud.

What you walk out with

  • A transcript of who said what
  • The questions asked, and the answers given
  • The decisions, and what was left open
  • The whiteboard, attached to the conversation it came from
  • All of it filed where you keep your notes, so tomorrow it knows today happened
The design goal: nothing from the meeting depends on someone's memory or someone's notebook.

Keeps the record

Listens the whole time. A rolling transcript while you talk, and a cleaner one afterward from a better recording, with nobody assigned to take notes.

Answers when asked

Tap the screen, or say its name, and ask. It answers from what was said in the room and from what you already know, your own notes and records, not just the open internet.

Stays yours

Strategy talk with partners is exactly what should not sit on someone else's transcription service. The audio and the transcript stay on a server you own.

How it would work

On the table: a small touchscreen board with a microphone and a speaker. It streams the room's audio over your network to your own server, which transcribes it as you go. When someone taps to ask, the question and the conversation so far go to a language model, Claude, and the answer comes back as speech and as text on the screen.

After the meeting, a better recording becomes the transcript you keep, with speakers separated and the photo of the whiteboard attached. Two paths on purpose: fast and rough for answering in the room, slow and careful for the record.

What it is not

It is not built. There is no device to buy and no waitlist. This page is the design, written down before the code, so the name means something when the code arrives.

And the limits, stated now rather than discovered later. It is for people in one room, not for Zoom or Teams calls. One microphone hears a table well and the far end of a boardroom badly. To answer a question, the question and the conversation so far do leave your server for the language model; the recording itself does not. And some places require everyone's consent to record, so Boswell announces itself when it starts.