A one-hour video call with a client has a moment that decides the outcome, and it's almost never the last one. It usually lands around minute twelve. The client asks what you billed them last year for the same thing, or whether what you promised in March ever shipped. You have the answer: it's in a spreadsheet, in an email or in the CRM, three clicks away. What you don't have are the fifteen seconds of silence it takes to go and find it without anyone noticing.
The whole AI notes market is built for minute sixty-one. Clean summary, decisions pulled out, tasks assigned, CRM record updated on its own. That's real work and they do it better than anyone would by hand. But minute twelve is still empty, and minute twelve is the one that was costing money.
01Minutes and live support are two different products
They're sold in the same category and they look alike in the demo, but they optimize for opposite things. A minutes tool can afford to take its time. It produces its artifact once the conversation is over, so it has the full transcript, it can reread the whole thing and it can spend two minutes writing something properly. The result gets shared, filed and searched six months later.
Live support has none of those comforts. It has only heard what's happened so far, it doesn't know how the sentence you're listening to ends, and its answer expires. A note that arrives late gets in the way more than it helps.
Why almost nobody covers that window
Because you have to listen to the call as it happens, cross what's being said with what's in front of you on screen —which is where the figure you're missing lives— and answer before the conversation moves on, all at the same time. Each of those on its own has been solved for years. The three at once, with latency that keeps up with human conversation, is a different engineering problem.
02What Granola and Otter do, and what they cost
Granola
Granola doesn't join the call as a participant: it listens from your computer and takes your notes as its base, improving them with what it heard. It's a good design, because the note that comes out looks like what you thought mattered and not what a model thought mattered. Its Basic plan is free with limited history, Business costs $14 per user per month and Enterprise, $35 per user per month.
On its pricing page, checked on 25 July 2026, there's no mention of voice conversation with the agent, of screen vision, of camera or of document generation. That's consistent with what they are: a notes tool, very well tuned. Third-party sources also report a 25-note cap on the free plan and availability on Mac and Windows with mobile on iOS only, but we could not confirm those three on their own site and we give them as approximate.
Otter.ai
Otter is the veteran of transcription. Its Basic plan is free and Pro runs around $16.99 a month, or about $8.33 a month billed annually, with roughly 1,200 minutes included. These figures come from third-party sources and are worth checking on otter.ai before you decide anything: transcription plans change often.
If you run meetings in a language other than English there's a wrinkle that tends to surface late. Otter has transcribed Spanish since 2024, but the interface, the support and the reports still work in English, according to users who have said so publicly. If you're the only one reading the notes, it makes no difference. If you're forwarding them to a client who doesn't read English, it starts to matter.
| What | Granola | Otter | Saelyx |
|---|---|---|---|
| Help while you're talking | Not advertised | Not advertised | Yes · floating panel |
| Summary when it's over | Yes, on your notes | Yes, on the transcript | Yes, emailed to you |
| Voice conversation with the agent | Not advertised | Not advertised | Yes |
| Sees your screen and your camera | Not advertised | Not advertised | Yes, both at once |
| Remembers the person from last time | Note history | Note history | Persistent memory and faces |
| Documents in Word, Excel or PowerPoint | Not advertised | Not advertised | Yes, dictated |
| Integrations and teams with SSO | Yes | Yes | No |
| Platforms | Mac and Windows (approx.) | Not checked | Mac |
| How to try it without paying | Basic, limited history | Basic | Invite-only; while it lasts, nothing is charged |
| Published price | $14 and $35 per user/month | ~$16.99/month (approx.) | Not yet |
03Where they win, no hedging
There are four areas where Granola and Otter are ahead, and it isn't a matter of nuance.
- Integrations. Having the note show up on its own in Notion, in a Slack channel, in a HubSpot record or chained through Zapier is half the value of the product. Saelyx has none of that.
- Teams. Single sign-on, control over who sees what, data governance, per-seat billing. They have it solved. Saelyx is a one-person product today, with no admin console.
- Shared history. Letting the colleague who wasn't there read Tuesday's meeting is a huge use case and it isn't ours.
- Windows. Half the desks in any office run Windows. Saelyx's Meeting Mode doesn't exist outside the Mac.
Put as uncomfortably as it can be put: if you're a twenty-person sales team that needs meetings to end up in the CRM, Granola is a better tool today than Saelyx, and buying from us would be a mistake.
04What Meeting Mode does
The scenario it's built for is narrow and very specific: one person, on their Mac, in a meeting that decides something. A negotiation, an interview, a committee meeting, a technical call with a client who knows their own system better than you do.
While the meeting is happening:
- It listens to the call and puts the answer, or the key point, in a floating panel only you can see. The panel refreshes every few seconds, not word by word.
- It doesn't speak, it doesn't introduce itself and it doesn't join the video call as a participant.
- It crosses what it hears with what it sees on your screen: the open spreadsheet, the email, last year's quote.
- It remembers the person in front of you from last time, and what you agreed on.
When you hang up:
- It leaves you the summary, with no watermark, and emails it to you if you ask.
- It takes down the minutes in Word or the quote in Excel, with their formatting, while you talk.
The speed part matters more than it looks. A note that lands when you've already moved on to something else is worse than getting nothing, because it distracts you. So it's worth separating two clocks that don't run at the same pace. When you talk to the agent directly, the engine starts answering in 776 ms measured in production. The meeting panel is slower: it goes back over what's been said every few seconds and drops in what it finds, so a suggestion can land with the next sentence rather than with that one. And if you ask for something that takes real work, reading a whole document or building a file, it goes quiet for a good while before coming back.
Meeting Mode, the memory that recognizes the person, live vision and dictated documents all live in the same Mac app, and today you get to it by invitation. There's no price yet: while access stays like this, putting a figure on it would be inventing one. When there is one, it will be published on pricing before anyone is charged a thing.
What Meeting Mode doesn't do
It doesn't push anything to Notion or Slack or the CRM, because we have no integrations. It doesn't produce a shared transcript for the team. There's no team version, no SSO, no admin console. It doesn't run on Windows, or on the iPhone: the iPhone and iPad versions are a target for the last quarter of 2026 and the dates can move. And it doesn't work offline.
One warning is missing, and it's about state rather than catalogue. Meeting Mode is the youngest feature in Saelyx and it's still being actively polished: on top of the pace of the panel, if the Mac goes to sleep with a meeting open you'll want to switch it back on before carrying on. That's why access is invite-only and we review every request by hand. With few users we can fix whatever turns up in days.
05Three moments where it shows
- Last year's price. The client asks what you charged them for the same line item. The spreadsheet is open on your second screen, just not on the right row. Instead of going quiet and searching, you ask under your breath or let it read, and the figure appears in the panel while the other person carries on talking.
- The objection you already know. A candidate or a client raises the same complaint someone else raised three weeks ago, the one you answered well back then. The agent remembers that conversation and hands you back the argument that worked, not a generic one.
- The minutes nobody writes. The meeting ends at fifty-eight past, the next one starts on the hour, and the minutes go unwritten for a week. With the summary in front of you, you dictate the decisions and a formatted document comes out before you join the next call.
06What to know before you ask for it
Saelyx runs on the Mac today, with macOS 13 or later. From any computer with a browser you can talk to the agent, but that's voice conversation: the floating panel over the video call is a Mac thing. The app is downloaded from our site, signed and notarized by Apple, not from the App Store.
On data, the line is the one that's published and we're not going to stretch it here. Your audio and your screen aren't stored in Saelyx's systems and aren't used to train models. There's one exception and it's worth reading twice: free access rides on a free engine where live data can be used. The raw transcript of a meeting is kept for 30 days; the summary, for as long as your account is active. The infrastructure sits in the European Union, in Ireland. You can export it all and delete the account from inside the app.
And access is invite-only. There's no open download: you ask, every request is reviewed by hand, and you get an answer. That makes the process slower than installing Granola in two minutes, and it's a real disadvantage we're taking on while the product grows carefully.
07Frequently asked questions
Does Saelyx replace Granola or Otter?
It depends what you use them for. If what you need is shared minutes, a history the whole team can search and meetings landing in Notion, Slack or the CRM on their own, then no: we don't have those integrations and they've spent years polishing theirs. If what you need is support while you talk, that's where Meeting Mode comes in, and neither of them advertises that.
Can the other people tell I'm using it?
The panel is yours and Saelyx doesn't touch the audio: it doesn't speak, it doesn't introduce itself and it doesn't join as a participant. It doesn't announce it for you either. Telling people or not is your call, and our recommendation is to tell them. There's nothing wrong with using live support; hiding it does complicate the conversation if it comes out later.
What happens to the audio and the transcript?
Audio and screen aren't stored and aren't used to train models, except on free access, which rides on a free engine where live data can be used. The raw transcript lasts 30 days and the summary stays as long as the account is active, with cascading deletion when you remove it. The details are in the privacy policy.
Does it work on Windows or on the iPhone?
Meeting Mode is Mac-only. Voice in the browser does work from Windows, Linux or a Chromebook, but that's conversation, not a panel over the video call. iPhone and iPad are a target for the last quarter of 2026.
How do I get it?
You ask for access at request access and it's reviewed by hand. Nothing is charged today and there's no published price; when there is one, it will appear on pricing before a charge reaches anyone.
If your problem is that meetings get lost, buy a minutes tool: there are two good ones up there and one of them is free. If your problem is that in the meeting you're half a second and one figure short of something you know you have, that's a different thing, and it's the one we're building.