Jamie: the meeting notetaker that never joins the meeting
The first time a notetaker bot slid into a client 1:1, I burned the opening two minutes explaining what the extra participant was. The second time, a client in Ostrava asked me to kick it out before he'd say anything real. A weekly one-on-one isn't the place for a third attendee with a record button, however well-behaved it claims to be. So the robot had to go, and I went looking for notes that didn't announce themselves in the attendee list.
Jamie's the one I stuck with.
Its defining trait, and the entire reason it fits my week, is that it doesn't send a bot into the call. It captures the meeting audio on my own machine, the system audio already coming out of my speakers, so there's no participant and no line that reads Notetaker has joined. That single architectural decision is the whole story of the tool, and it's the axis I pull apart in a bot in the call versus audio on your device.
what it got right for me
Two things, mostly.
First, the silence. Meet, Zoom, Teams, Jamie doesn't care which, because it listens to what my Mac is already playing rather than logging in as a guest. Clients see me and only me. For the accounts where a visible recording bot reads as faintly adversarial, that counts for more than any checkbox on a comparison grid.
Second, and this genuinely surprised me, it caught an in-person meeting. I sat in a cafe in Brno with a founder, no video call at all, laptop open on the table, and got a structured summary with action items afterward exactly as I would from a remote call. A bot can't do that, because there's no meeting for it to join. My own local Whisper pipeline can, but only because I wired the audio routing by hand; here it simply worked.
the german angle a client actually cared about
Jamie comes out of a Berlin company, and as of this writing it leans hard on an EU-data, GDPR-conscious position. For most of my work that's a pleasant footnote. For one Brno client with firm views on where their words get processed, it was the hinge the whole decision turned on. They didn't want meeting content sitting on a US server under retention terms drafted three time zones away, and being able to point at a European vendor with a straight face ended the debate. I still told them to read the actual data-processing terms rather than trust me, which is the entire spirit of where your meeting audio goes.
A notetaker that never appears in the attendee list is a different social object than a robot that does, and people feel that difference before they can put words to it.
where it bit me
The on-device model carries a tax, and it's the obvious one the moment you say it aloud: the machine has to be present, awake, and running the capture. My Mac sleeps, or I shut the lid to walk to a meeting room, and a gap opens in the audio that the summary quietly papers over. A bot living in the cloud doesn't doze off when I do. I trusted the thing for the better part of a month before I noticed that two sets of notes had swallowed a five-minute stretch I'd stepped away for, with nothing flagging the hole.
The other edge is the one I hit no matter whose software I run: heavy accents. Czech-accented English, mine very much included, is where transcription wobbles, and a summary is never better than the transcript beneath it. Jamie handled it better than I feared and worse than a native-English room would land, the same failure family I documented on my own rig, just someone else's model wearing it.
So I run both, Jamie for the ordinary calls where I want notes without ceremony, my offline pipeline for the NDA work where custody isn't up for discussion. The homemade rig stayed. What left was the robot, and that's all I actually wanted.