There is a specific, universal moment in meetings: you surface from a thought, realise a question has been asked, and have no idea what the last two minutes contained. You have three options, all bad. Ask them to repeat it and admit you drifted. Stay quiet and hope context arrives. Or guess, and be wrong in front of the client.

The same moment happens for different reasons all day. You joined ten minutes late because the previous call overran. You are the only person in a time zone where this call is at 11pm. You are on your second back-to-back client conversation and the names have started to blur. In every case the information exists — it was said out loud, thirty seconds ago — and you have no way to reach it without costing someone else their time.

Why the transcript alone does not solve this

Most AI meeting tools will happily give you a transcript, and afterwards that is useful. During the call it is close to useless. Scrolling a live transcript to find the sentence you missed means reading while people keep talking, which is how you miss the next two minutes as well. You have traded one gap for a bigger one.

What you actually want is not the text. It is the answer. "What did they just decide?" is a question, and the transcript is only raw material for it.

Asking the meeting a question

Lunar keeps a live in-meeting chat that answers from the transcript as it is being written. You type a question and get an answer, silently, while the call continues.

The questions people actually ask fall into a few shapes:

Recap requests. What did they say in the last five minutes? You get a short summary of what you missed rather than a wall of text to read.

Commitment checks. What have I agreed to so far? Genuinely useful near the end of a long call, when you are about to say yes to a fourth thing.

Topic checks. Did anyone mention the migration? Faster than scanning, and it works when you half-remember a subject but not a phrase.

Decision checks. What did we decide about the deadline? The one you need most on client calls, where getting it wrong is expensive.

Because you ask privately rather than in a shared document, nobody in the meeting knows you asked — which is the entire point. This is the difference between a live collaborative transcript, where the notes are a shared artefact everyone can see, and a private assistant that answers only to you.

Where this matters most

Distributed teams. If half your team joins calls outside their working hours, the ability to arrive late and catch up in fifteen seconds is not a convenience, it is what makes the meeting survivable. We wrote more about that in async communication.

Back-to-back client work. Consultants and agencies switch context more times per day than anyone. Reloading who said what, without asking the client to repeat themselves, protects the impression that you are on top of the account — which you are, you just have six of them.

Long technical calls. In a two-hour architecture discussion, nobody holds the whole thread. Being able to check what was concluded about the database an hour ago, without derailing the current topic, keeps the conversation moving.

Meetings you should not be in. Sometimes you are on a call for the one part that concerns you. Asking whether your area has come up yet is a reasonable way to stay present without staying fully occupied.

How it works, technically

Lunar is a desktop app. It captures your microphone and your computer's system audio directly, so no bot joins the call and nothing appears in the participant list on Zoom, Google Meet, Teams, or anywhere else. The transcript builds locally as the meeting runs, and the in-meeting chat answers against that transcript.

Because capture is happening on your machine rather than through a service that joined the call, live answers work on any platform without an integration, and without anyone else on the call knowing a tool is present. If you want the fuller argument for bot-free capture, it is in recording meetings without a bot.

What it will not do

It answers from what was said, so it cannot tell you what someone meant, and it cannot resolve a genuine ambiguity in the conversation. If two people talked past each other about a deadline, the transcript contains that confusion and so will the answer.

It also will not save you from a meeting that needed you to be present. Catching up on two minutes is a repair for a normal lapse in attention. Catching up on forty is a sign you did not need to be on the call — which is worth noticing rather than solving with software.

Frequently asked questions

Can other people see that I asked something?

No. The chat is private to you and local to your machine. Nothing is posted into the meeting, and no shared document changes.

Does this work if I join a call late?

Yes, provided Lunar was running when the meeting started — it notices calls starting and offers to record, so in practice it captures from the beginning even when you arrive halfway through. You can then ask for a recap of everything before you joined.

Does it work on Zoom, Meet and Teams?

All of them, and anything else that makes sound. Because capture happens at the system audio level rather than through a per-platform integration, the platform does not matter.

Is my audio being sent somewhere while the meeting runs?

Sessions stay on your machine until you sign in, and transcription can run fully on-device if you choose. Live chat answers use cloud AI today, and we say so plainly rather than burying it — the security page shows the exact data flow per feature.

What happens after the meeting ends?

You get a summary of decisions and action items, and the action items can become Jira issues in the right project. That side of it is covered in turning meeting action items into Jira issues.

The short version

The information you missed already exists. The only real question is whether reaching it costs somebody else their attention. Being able to ask quietly, and get an answer in a couple of seconds, removes one of the small daily frictions that nobody ever puts on a roadmap — and that everybody has.