Key takeaways

• The free plan: 40 minutes per meeting, 5 participants, one meeting at a time. No recording, no AI summaries, no whiteboard, no booking pages.
• The 40-minute limit is per meeting, not per day. Starting a new meeting resets it — the workaround everyone discovers is not against the rules.
• "5 participants" counts everyone in the room, you included. Four guests, not five.
• What is *not* limited: video quality, screen sharing, chat, polls, Q&A, the lobby and password controls, or how many meetings you have per month.
• Everything paid starts at Pro. Current prices in rupees are on the pricing page — this post deliberately does not quote them.

Every video platform has a free tier, they all sound roughly the same in the marketing, and the differences only surface at the worst possible moment — usually about 39 minutes into something that mattered. This post is the boring, specific version for RoundMeet: what the free plan gives you, what it does not, and what actually happens when you hit a limit.

The limits, precisely#

Three of those rows deserve more than a cell.

40 minutes, per meeting#

The cap is on a single meeting, not on your day or your month. When a meeting reaches it, the meeting ends — and starting a new one begins a fresh 40 minutes.

This is worth saying plainly because every free-tier user eventually discovers it and then wonders whether they are cheating. They are not. It is a limit on continuous meeting length, and rejoining is the documented behaviour, not a loophole. It is also genuinely annoying in the middle of a workshop, which is the honest reason the limit exists at all.

Practically: if you know a call will run long, tell people at the start that you will re-send the link at the 40-minute mark. A planned break is fine. An unplanned disconnection during a client's sentence is not.

5 participants, including you#

The number is everyone in the room. You plus four others.

The sixth person cannot join — they are refused at the door rather than dropped mid-meeting, which at least means the failure happens to the person arriving rather than to the meeting. For a one-to-one, a small team stand-up or a tutorial, five is genuinely enough. For anything class-shaped it is the limit that bites first, well before the 40 minutes does.

One meeting at a time#

You cannot host two meetings simultaneously on the free plan. This only matters if you are a team sharing one account, which is a thing people try, and it is the point at which it stops working.

What the free plan does *not* limit#

This is the part most comparison articles skip, and it is where the real answer to "is the free tier usable?" lives.

• Video quality. Free meetings are not degraded. Same resolution, same media path, same bridge.
• Screen sharing. Full, including sharing a specific window or tab.
• Chat and file sharing in the meeting.
• Polls and Q&A. Both work — see polls and Q&A.
• Security controls. The access token, the password lock and the lobby are all available. Nothing about keeping people out of your meeting is behind a paywall, which is a line worth being deliberate about.
• Number of meetings. There is no monthly quota. Forty 40-minute meetings a month is fine.
• Breakout rooms, hand raise, virtual layouts, low-bandwidth mode, Car Mode.

So the free plan is not a demo with the good parts removed. It is the whole meeting experience with a stopwatch and a small room.

What is actually behind the paywall#

Four things, and they cluster more neatly than the table suggests:

Recording, transcripts and AI summaries. These are the expensive ones to run — a recorder process joins your meeting, then a transcription model runs over the audio, then the file is stored for a month or a year. That is real compute and real disk per meeting, which is why they start at Pro rather than being metered on free. How recording works covers what you get.

Room size and meeting length. The 40-minute and five-person caps.

The whiteboard.

Booking pages. Zero on free, three on Pro, fifty on Business. If the reason you are here is letting people book time with you, that needs Pro.

Is the free plan enough for you?#

It comes down to two questions, in this order.

How many people? Six or more in a room, regularly, and no arrangement of the other settings helps. This is the hard wall.

Do you need the meeting to still exist afterwards? If your meetings need a recording, a transcript or a summary — training sessions, interviews, anything with minutes — free will not do it, and no amount of rejoining works around that.

If neither applies, free is a real answer rather than a trial. One-to-ones, pair work, small stand-ups and tutorials all fit comfortably inside it.

Prices#

Deliberately not in this post. Prices change, blog posts do not, and a stale number is worse than no number: the pricing page has the current figures in rupees, including what GST does to them.

What is stable enough to say here is the shape. Pricing is per user per month, billed to the account that hosts the meeting — the people who *join* your meetings are never charged and never need an account, so a five-person team pays for five people, not for everyone they ever meet with.

The short version#

Free is 40 minutes, 5 people, one meeting at a time, and no recording — with everything else, including all the security controls, fully working. If your meetings are small and you do not need them to exist afterwards, that is a real plan. If you need six people or a recording, it is not, and no setting changes that.

Key takeaways

• Only the host can start a poll or open Q&A. Everyone can vote, ask and upvote once it is running.
• A poll is one question, up to ten options, live results, and one vote per person enforced by the server.
• Q&A sorts by upvotes, not by time. The room decides what gets answered, which is the whole point.
• Q&A must be opened by the host first — until then participants see *"The host hasn't started it yet."*
• Both survive a refresh. Someone who reloads or drops off and rejoins gets the current state back, votes included.

Somewhere past about a dozen participants, a meeting stops being a conversation. Unmuting becomes an act of courage, the same three people answer everything, and "any questions?" reliably produces four seconds of silence followed by somebody moving on.

Polls and Q&A are both attempts at the same fix: lower the cost of participating from *speaking in front of thirty people* to *typing*. They solve different halves of it, though, and using the wrong one is why these features so often sit unopened.

Polls: when you want a number#

A poll answers a question you already know how to ask. Use it to make a decision, to check that people are still following, or to turn a vague disagreement into a count.

Starting one#

Open the Polls panel from the control bar. As the host you get the creator; type the question, fill in the options, and start it.

• One question, up to ten options.
• The question can run to about 300 characters, each option to about 200 — enough for a real sentence, short enough that nobody writes a paragraph into a radio button.
• Two options minimum. You can remove options down to two and add up to ten.

Everyone in the meeting sees it appear immediately, host included. Non-hosts who open the panel before there is a poll see *"No active poll right now."*

While it is running#

Results update live as votes come in. One vote per person, enforced on the server rather than in the browser, so refreshing the page or rejoining does not buy anyone a second vote.

Two controls belong to you:

• Close voting freezes the result. The poll stays on screen with its numbers and a *Closed* label, and nobody can change their answer. This is the one to use — the count stops moving while you talk about it.
• End poll removes it from the meeting entirely.

What polls are not good for#

Anything open-ended. A poll needs you to have written the options, which means you have already decided what the possible answers are. If you do not know them yet, you want the other tool.

Q&A: when you want the questions themselves#

Q&A is a queue that the room sorts. Questions arrive, everyone upvotes the ones they also wanted asked, and the list orders itself by votes rather than by who typed first.

That ordering is the entire feature. In a chat panel the loudest and fastest question wins; in Q&A the *most widely shared* one does — including the one four people were quietly wondering about and none of them would have unmuted to ask.

Opening it#

Q&A starts closed. Open the Q&A panel and, as host, press Start Q&A. Until you do, participants see *"Q&A isn't open yet — the host hasn't started it yet."*

Making it explicit is deliberate: an always-open question queue collects questions during the part of the session where you are not taking any, and then you are answering things from twenty minutes ago.

How it works once open#

• Anyone types a question — up to about 500 characters, so a question with context in it fits.
• Everyone else can upvote. One upvote per person, and you cannot take it back.
• The list re-sorts by upvotes as votes land.
• Marking a question answered greys it out and strikes it through. It stays visible, so nobody asks it again and latecomers can see it was covered.
• A question can be dismissed by the host, or by the person who asked it.

Questions carry the asker's name. That is worth knowing before you promise anonymity — this is a named queue, not an anonymous one.

Which one, when#

Running them well#

Open Q&A at the start, take questions at the end. Say so. People think of questions during the talk and forget them by the end; the queue is where they park them.

Read the top question out loud before answering it. Half the room is not looking at the panel.

Mark things answered as you go. It is two seconds and it is the difference between a queue and a list.

Do not leave a poll open all meeting. Close voting once the numbers stop moving. An open poll pulls attention away from whoever is speaking.

In a class, poll more often than feels necessary. A poll every ten minutes is the cheapest attendance-and-attention signal there is, and students answer polls they would never unmute for. Q&A is where the questions they are embarrassed to ask actually show up.

The short version#

Polls answer a question you have already written. Q&A finds out which questions the room actually has, and sorts them by how many people share them. Open Q&A early, take from it late, and close a poll once you have your number.

Key takeaways

• A meeting link is just a URL. Anyone who has it can try the door, so "nobody will find it" is not a security control.
• Three controls exist, and they stack: the access token in your invite, a password lock on the room, and a lobby that holds each person until you admit them by name.
• Only the host sees these. Everyone else is told, honestly, that the meeting's security is managed by the host.
• The lobby is the only one that lets you see who is asking before you decide. Use it for anything with outsiders in it.
• Locking a room affects people who join *after* you lock it. It does not remove anyone already inside.

The uncomfortable thing about a video meeting link is that it is a URL. It gets pasted into a group chat, forwarded to somebody who could not make it, and copied into a calendar invite that syncs to three other calendars. None of that is a breach; it is what links are for. It just means "the link is private" cannot be the whole plan.

RoundMeet has three controls for this, and they answer three genuinely different questions. This post is about which one to reach for.

Where the controls live#

All three are in one place. In a meeting, open the More (⋮) menu in the control bar and choose Security.

If you are not the host, that panel says so — *"This meeting's security is managed by the host"* — and shows nothing else. That is deliberate. A participant who can see which protections are off knows exactly which door to try.

1. The access token: proof you were invited#

Every meeting has an access token, shown at the top of the Security panel with a Copy button next to it.

It is not a password you type at a prompt. It is the part of the invite that says *this person was invited*, and it travels with the link you share. Its job is to stop someone who guessed or scraped a room name from walking in.

That makes it a good default and a weak boundary. It protects against a stranger who never had the invite; it does nothing at all once the invite itself has been forwarded — and the whole reason the other two controls exist is that invites get forwarded constantly, usually with the best intentions.

Use the Copy button rather than retyping it. It is the most commonly mistyped thing in the panel.

2. The password lock: a shared secret at the door#

In the Security panel, Password lock has a text field and a Lock button. Type a password, press Lock, and the room is locked; the panel then reads *"This room is locked with a password"* and offers Remove password instead.

Anyone joining from that point on has to enter it. That is the important qualifier:

A password is the right control when you can get the secret to the right people over a *different channel* from the link. Sent in the same message as the link, it adds a step and no security whatsoever: whoever forwards one forwards both.

Two practical notes. Set the password before you announce the meeting, not in the first minute, or the early arrivals will be locked out by a lock you added while they were joining. And pick something you can say out loud on a phone call — a password that has to be spelled character by character over a bad line costs more time than it saves.

3. The lobby: you approve each person by name#

Lobby (waiting room) is a single toggle. Turn it on and nobody joins directly any more. They land in a waiting state, and their name appears in your Security panel with two buttons: Admit and Deny.

This is the only one of the three that lets you *see who is asking before you decide*. The other two are locks — the right key opens them and there is no judgement involved. The lobby is a door with a person behind it.

That makes it the correct choice whenever the meeting has people from outside your organisation in it: an interview, a client call, a class where you know the roll, anything where a name you do not recognise is itself the signal. It is also the only control that helps against the one failure the others cannot touch — a link forwarded to someone who *was* trusted and no longer is.

The cost is real, though: someone has to be watching the panel. If you are presenting to forty people, admitting them one at a time while you talk is not workable, and the latecomer knocking at minute twenty will sit there. For large sessions, prefer the password, or turn the lobby off once the meeting has properly started.

Which one should you use?#

They stack, and stacking is normal. A token plus a lobby is the everyday combination for meetings with outsiders in them. All three at once is not paranoid, just slow.

The part no control covers#

Everything above governs *joining*. Once someone is in the room, the questions are different, and they have different answers:

• Someone is in and should not be. Open the Participants panel and remove them. Then lock the room so they cannot walk back in.
• You are recording. Say so out loud. Everyone sees the red REC badge and hears the chime, but the badge is a notification, not consent — see how recording works.
• Someone is disrupting the meeting. Mute them from the Participants panel; that is faster than removing them and usually enough.

The short version#

The access token proves an invite. A password proves a shared secret. The lobby proves you looked at the name. Pick the one that matches what you are actually worried about — and if you are worried about a forwarded link, only the lobby helps.

Key takeaways

• A booking page lives at /book/<your-handle> and shows only slots that are genuinely free — it checks your existing meetings before offering anything.
• You control slot length, buffers between meetings, how much notice you need, and how far ahead people can book.
• Availability is set as weekly windows in your own timezone, so a 9:00 slot stays 9:00 through a DST change.
• Booking creates a real scheduled meeting. The guest and you both get an invite with a calendar file and reminders.
• You can put the meeting in your personal room, or have each booking get its own room. The default — a fresh room each time — is usually the right one.

The scheduling email is a small tax that everybody pays. Three messages to find a time, one to confirm it, and one more when somebody realises it clashes. For anything you do repeatedly — office hours, demos, interviews, consultations — a booking page removes it entirely.

What a booking page actually is#

It is a public URL: /book/ followed by a handle you choose. Anyone with that link sees a page with your name, a short description, and a set of times you are free. They pick one, enter their name and email, and it is booked.

The important word is *free*. The page does not show a static timetable. Every time it loads, it works out your open slots by taking your weekly availability, cutting it into slots, and then removing everything that collides with a meeting already on your RoundMeet schedule — including recurring ones. A slot that shows up is a slot you can actually take.

Setting one up#

1. Pick a handle#

This becomes the public URL, so pick something you would be happy putting in an email signature. Short and human beats clever.

2. Set your weekly availability#

Availability is expressed as windows on days of the week — Monday 10:00–13:00, Tuesday 15:00–18:00, and so on. You can have several windows on the same day, which is how you protect a lunch break: two windows with a gap, not one long one.

These are in your timezone, and that is stored on the page rather than being inferred. It is what makes a 9:00 slot stay at 9:00 when clocks change at either end — if the windows were stored in UTC, half your slots would drift by an hour twice a year.

3. Set the four numbers#

Two of these are worth more thought than they usually get.

Minimum notice is the one that protects your morning. At the default of two hours, nobody can book the 9:00 slot at 8:45. Set it to a full day if you need to prepare for whatever the meeting is; leave it short if the point of the page is to be reachable.

Buffer is the one people skip and then regret. With a zero buffer, back-to-back bookings are genuinely back-to-back: the second meeting starts the second the first one ends, on a page that will happily book you six of them in a row. Ten or fifteen minutes is usually the difference between a useful day and an unbroken one.

Maximum advance mostly protects against being committed to something three months out that you have entirely forgotten about by the time it arrives.

4. Decide which room it uses#

Two options:

• A new room each booking (the default). Every meeting gets its own room code. Two bookings can never collide, and a link from a past meeting is dead.
• Your personal room. Every booking uses the same room. Convenient if it is always you and the link never changes — but if a meeting runs long, the next guest arrives in the middle of it.

Unless you have a reason, take the default.

5. Turn on the protections you want#

The booking page can pre-arm the meeting it creates: require the access token, or switch the lobby on so you approve each guest when they arrive. For anything with strangers in it — and a public booking page is by definition open to strangers — the lobby is the one worth having. See three ways to keep strangers out of your meeting for what each control does.

What happens when someone books#

A real scheduled meeting is created — the same kind you would have made yourself. The title combines your page's title with the guest's name, so your schedule reads *Office hours — Priya Sharma* rather than a wall of identical entries.

Both sides are invited: the guest gets an email invite with a calendar attachment they can add in one click, and you are invited to your own meeting so it lands in your schedule and you get the reminders. The slot then disappears from the page, because the collision check that generated it now sees a meeting there.

Making it work in practice#

Give the page a description that sets expectations. "30-minute intro call — tell me what you're building" gets better meetings than a blank page. The description is the only briefing your guest gets.

Do not offer all your working hours. A page offering 9-to-6 five days a week will fill in with things that could have been email. Offer the hours you actually want to spend this way.

Use separate pages for different things. A 15-minute intro call and a 60-minute review want different slot lengths, different notice periods and different descriptions. Two handles is cheaper than one page that fits neither.

Check the timezone before you share it. It is the setting people get wrong, and the failure is silent: everything works, everyone shows up an hour off.

The short version#

Set your weekly windows in your own timezone, give yourself a buffer and a real notice period, take the default of a fresh room per booking, and turn the lobby on. Then send the link instead of the email.

Key takeaways

• Most "bad call" symptoms are one cause: you are receiving more video than your connection can carry. Your upload is rarely the problem — your download is.
• Low bandwidth mode caps you at three incoming video streams and drops every one of them to 180p. It is in the More (⋮) menu and it works immediately.
• Car Mode turns incoming video off entirely — audio only, no streams at all. It is the strongest lever in the product.
• Turning off *your own* camera helps everyone else a little and you almost not at all. Turning off *incoming* video is what fixes your call.
• Wi-Fi at the far end of the house is a more common cause than the internet connection you are paying for.

Ask ten people to describe a bad video call and you get ten descriptions: video that freezes into a still frame, audio that turns robotic, a green flicker across somebody's face, echo, being told repeatedly that you are breaking up, and a call that quietly drops every few minutes. They sound like different faults. Almost always they are the same one.

The one cause behind most of it#

In a group call you are not sending one video stream. You are *receiving* several — one per person whose camera is on — and each of those is a continuous download that never pauses.

That is the asymmetry people get wrong. Your camera sends a single stream, and modern codecs are good at squeezing it. But a five-person call means four incoming streams, and unlike a file download they cannot slow down and arrive late: a video frame that arrives late is a video frame that is thrown away. When the pipe is not wide enough, the software starts discarding, and *discarding is what all those symptoms are*. Frozen video is discarded frames. Robotic audio is discarded audio packets. The dropped call is the connection giving up after discarding too much.

Which is why the first thing everyone tries — turning off their own camera — is close to useless for their own experience. It helps the *other* people slightly. It does approximately nothing for you, because your problem is what is coming in.

What to do, in the order that helps most#

1. Turn on Low bandwidth mode#

Open More (⋮) in the control bar and choose Low bandwidth mode.

This is the single most effective control in the product, and it is worth knowing exactly what it does rather than treating it as a vague "try harder" switch. It makes two changes at once:

• You receive video from at most three people — whoever is currently active — instead of everyone.
• Every stream you receive, and the one you send, is capped at 180p instead of the usual 720p.

Those two together cut your incoming data by roughly an order of magnitude in a busy room. Faces get soft. That is the trade, and in a call that is currently freezing it is not close: 180p that keeps moving beats 720p that stops.

It takes effect immediately, it applies only to you, and nobody else can tell you have turned it on.

2. If that is not enough, use Car Mode#

Also in the More (⋮) menu. Car Mode receives no video at all — zero streams — and turns the call into audio with a simple overlay showing who is speaking and how long you have been in the meeting.

The name comes from its original purpose, but the mechanism is what matters here: if audio is still breaking up in Low bandwidth mode, there is not enough connection for any video, and continuing to receive some is what is breaking the audio. Car Mode is not a lesser experience with video removed; it is the experience that works when nothing else does.

If you need to see a shared screen, this is not the mode for you — Car Mode drops incoming video, including the screen share.

3. Move closer to the router, or use mobile data#

This is unglamorous and it is the most common actual fix.

A 100 Mbps connection is a fact about the cable outside your house. What reaches your laptop at the far end of two walls may be a fraction of it, and — this matters more than raw speed — much less *steady*, which is what real-time video actually needs. Video calls tolerate a slow, consistent connection far better than a fast, erratic one.

So before you blame the platform: stand next to the router and rejoin. If the call is suddenly fine, the internet was never the problem. Wired Ethernet, if you have it, removes the variable entirely.

Mobile data is often better than weak Wi-Fi for exactly this reason. A 4G connection with two bars is usually steadier than Wi-Fi at the edge of its range.

4. Close the other tabs#

Video decoding is CPU work, and a laptop that is thermally throttling drops frames in a way that looks identical to a network problem. Twenty Chrome tabs, a video playing in one of them, and a build running in the background will each take their cut.

If your fan is loud and the video is choppy, try the laptop-side fix before the network-side one.

5. Check the actual numbers#

More (⋮) → Meeting Statistics shows what is really happening rather than what it feels like. Look at packet loss and the connection quality indicator for each participant.

The useful thing here is that it tells you *whose* problem it is. If everyone in the room looks bad to you but you look fine to everyone else, it is your download. If one person looks bad to everybody, it is theirs — and no amount of you switching modes will fix it.

Symptom to cause#

What does not help as much as people think#

• Turning your own camera off. Helps the room a little, helps you almost not at all.
• Rejoining. Sometimes it works, because it re-picks a media path. Usually you land in the same place. Try it once, not four times.
• Switching browsers. Modern Chrome, Edge, Firefox and Safari all handle this comparably. It is worth a try only if one specific browser is misbehaving for you generally.
• "Turning off HD" in some other app. Video calls are not competing with your settings; they are competing with your neighbours' streaming and your own downloads.

The short version#

Your download is the bottleneck, not your upload. Turn on Low bandwidth mode first, drop to Car Mode if audio is still breaking up, and move closer to the router before you blame anything else. Turning off your own camera is the fix everyone tries and the one that helps you least.

Key takeaways

• Recording is server-side. RoundMeet's recorder joins the call and captures a single MP4 on our infrastructure — your browser records nothing and your upload bandwidth is not used.
• Only a host can start it, and only on a plan that includes cloud recording. Everyone in the meeting sees a red REC badge and hears a chime the moment capture actually begins.
• The file does not land in your Downloads folder. It is filed under your account and appears on the Recordings page, moving through Queued, Transcribing, Summarizing and then Ready.
• A finished recording gives you the video, a timestamped transcript whose timestamps seek the player, and an AI summary with action items and key topics.
• Recordings count against your plan's storage. Deleting one removes its transcript and summary with it and gives the space back.

Every meeting platform has a Record button and every one of them puts the resulting file somewhere slightly different. This is the specific answer for RoundMeet: who is allowed to press it, what the other people in the call experience when you do, where the video ends up, and what happens to it in the minutes after everyone hangs up.

Who can record, and on which plan#

Two things have to be true before the Record button does anything. You have to be the host — the moderator, which normally means the person whose account created the meeting — and your account has to be on a plan that includes cloud recording.

If you are not the host, the button is there but disabled, and its label reads *Record (host only)*. On the Free plan it is disabled for everyone including the host, and the label points at an upgrade instead. Nothing about this is hidden: you can always see whether recording is possible in this room before the meeting starts.

Tying recording to the host is deliberate. A recording is the one meeting artefact that outlives the meeting, so the ability to create one belongs to the person accountable for the room rather than to whoever happens to be in it.

How to record a meeting in RoundMeet#

1. Join the meeting as the host#

Start the meeting from your own dashboard, or from a scheduled entry you own. The Record button reads your moderator role, so joining somebody else's link as a guest will not unlock it no matter what plan you are on.

2. Press Record#

It is the filled-circle icon in the control bar at the bottom of the call, near the More (⋮) button. On a phone the control bar scrolls sideways — swipe the row of icons if you cannot see it. It is in the bar, not tucked inside the More menu.

3. Wait about ten to thirty seconds#

Pressing Record does not start capturing instantly, and this is the step people get wrong. RoundMeet spins up a dedicated recorder that joins your meeting the way a participant does and captures what it sees and hears. While it is joining, the button shows a spinner and an amber Processing… chip appears in the meeting header.

When capture genuinely begins, the chip becomes a red REC badge that every participant can see and everyone hears a short three-note chime. Those two signals are the truth; the button turning red is not. If you are recording something that starts on the dot — a class, a webinar, a formal review — press Record before the first sentence, not on it.

4. Stop when you are done#

Press the same button again; while recording it is red and its label reads *Stop recording*. The recording also ends when the meeting does, so forgetting to stop it costs you a few seconds of an empty room, not the file. There is no save step and no file dialog — nothing is written to your computer at any point.

Where the recording file goes#

Nowhere near your laptop. Because capture happens on the server, the recorder writes a single MP4 — one mixed video of the meeting as a participant in the room would have seen it, with the audio of everyone who was unmuted — and hands it to RoundMeet, which files it under the account that owned the room.

You will find it on the Recordings page, also reachable from the quick actions on your dashboard. A finished recording usually appears there within a minute or two of the meeting ending, then works through a short pipeline. The status pill on each row tells you where it is:

• Queued — the file has arrived and is waiting for a processing worker.
• Transcribing — the audio is being turned into text.
• Summarizing — that text is being condensed into a summary and action items.
• Ready — everything finished. Only Ready rows open.

A row marked Failed means processing stopped before the transcript and summary were written. The meeting itself was still captured; what failed is the text layer on top of it.

What a finished recording gives you#

Open a Ready recording and there are three things on one page:

• The video, playing inline with normal seeking. The Download button beside it saves the MP4.
• The transcript, split into timestamped segments. Every timestamp is a button: click 04:12 and the player jumps to four minutes twelve. This is far and away the fastest way to find the one exchange you actually came back for.
• The AI summary — a paragraph on what the meeting was about, a list of action items with an owner attached wherever somebody was clearly named, and the key topics as tags.

What the AI summary does — and what it does not#

The summary is produced in two steps, and the difference between them matters if you care where your audio goes. The transcript is generated on RoundMeet's own servers using an open-source speech-to-text model; the audio is not sent to a third party to be transcribed. The written summary and the action items are then produced from that text by an external language model.

The summary is also the least important step, and it is built that way on purpose. If it cannot run, the recording still becomes Ready with the video and the transcript intact and the page simply says the summary is unavailable. A recording is never lost because a summary failed.

Treat the output as notes, not minutes. Summaries paraphrase, they occasionally attach an action item to the wrong person, and they cannot tell that the important moment in the meeting was a fifteen-second aside. The transcript is the record; the summary is the shortcut.

Storage, deleting, and what counts against your plan#

Every recording counts against your plan's recording storage. When an account runs out, the Record button switches off until space is freed — the platform refuses to start a recording it cannot store rather than accepting one and truncating it later.

Deleting a recording from the Recordings page removes the video, its transcript and its summary together and returns the space. There is no separate step for the transcript and no copy left behind.

Cloud recording and AI summaries start at Pro. Current prices in rupees live on the pricing page — a price is the one number a blog post should not hard-code.

Questions people actually ask#

The short version#

Be the host, be on a plan with recording, press the circle in the control bar, wait for the red REC badge, and stop when you are done. The MP4, the transcript and the summary are waiting on your Recordings page a few minutes later.