Should I vibe code
Bot-free meeting recording, notes, clips, and team knowledge
Removing the bot from the call also removed the only thing that told people they were being recorded.
?
Their verdict, the Pro price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-04.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
A Chrome extension that captures the current tab, an upload, a Whisper call and a summarisation prompt is a weekend, and that is precisely the problem — this is easy enough that you will actually do it. What Bluedot sells is not the transcript pipeline, it is the absence of a bot, and the absence of a bot is a design decision with a legal shadow. A recorder that joins the call announces itself; a browser extension does not, and in a two-party-consent state or under GDPR the notice is not optional just because your architecture removed the natural place to put it. Then two more things stack on top. It captures video, which means it also captures whatever anyone screen-shares: a spreadsheet of somebody else's customers, a dashboard, an inbox. And it is a team product, so the recording of a difficult one-to-one and the recording of a candidate interview end up in the same searchable workspace, one authorisation check away from everyone.
What actually breaks
not "if". the specific failures.
- Consent, structurally: there is no bot in the participant list, so unless you built a notice, nothing tells the other seven people on the call that this one is being kept
- Screen shares, which the video capture takes along with everything else — the shared tab with another customer's data in it is now in your transcript store and your search index
- Share links, where "anyone with the link" is how a recording of a performance conversation ends up somewhere it was never meant to be, and unguessable is not the same as private
- The workspace boundary, once recordings are a team asset: one missing per-object check and a searchable archive of every private conversation in the company is browsable by everyone in it
- Interview recordings, which are hiring records with their own rules about retention, access and what a candidate can ask to see
- The extension's permissions, because tab capture plus broad host access is a lot of trust to ask a whole team for, and the update channel is you
- Diarisation, which is the feature that makes a transcript useful and the one that confidently attributes a commitment to the wrong person
- Meeting platforms, which change their DOM and their media handling without notice, so the capture that worked in March silently records sixty minutes of silence in April
- Retention, which defaults to forever because nobody chose otherwise, and forever is the answer you will have to give when somebody's lawyer asks what you still hold
The message comes from someone two levels up and is only one line long: does the recording of the September call still exist? It does. So does the video, which means so does the eleven minutes where the other side screen-shared a spreadsheet with three of their own clients' figures on it — data you were never given, cannot lawfully retain, and have now had indexed for search and summarised into a note that six people can read. You go looking for who has opened it. There is no access log, because logging reads felt like paranoia when the whole thing was a side project. You go looking for the consent notice you were sure you added, and find that it renders in the extension popup, which is a thing nobody opens, and that the entire premise of the product was that participants would never see a bot in the list. The recording cannot be undone. What is left is deciding, in writing, whether to tell the other company that you have their clients' numbers.
Is that you?
the verdict is a default, not a law
- Every call is one you are on alone or with people who have agreed in writing, in advance, each time
- It records audio only, never video, so a screen share cannot end up in your archive
- The files stay on your own machine and there is no sharing feature at all
- You have a retention period, it is short, and something actually deletes on it
- Anyone on the call is outside your company, or in a jurisdiction where all parties must consent
- Recordings are shared to a team workspace, or reachable by a link that does not check who is asking
- You are recording job interviews, performance conversations or anything with an HR shape
- Video capture is on and you have not thought about screen shares
- Nothing tells participants they are being recorded, before recording starts
If you build it anyway
the checklist, then the prompt that enforces it
- Build the notice before the recorder. Announce in the meeting chat when capture starts, keep an on-screen indicator, and make stopping obvious — the whole point of the bot you removed was that it was visible.
- Default to audio only. Video capture is what pulls a screen-shared spreadsheet full of someone else's data into your archive, and once it is in the search index it is a disclosure problem rather than a storage one.
- Check permissions on every read of a recording, transcript and clip, against the requester rather than against the URL. A share link is not authorisation; it is a string somebody can forward.
- Set retention before you set anything else, make it a number in a config file, and prove that deletion reaches the media file, the transcript, the summary, the search index, the clips and any CRM record you pushed it into.
- Log reads, not just writes. The question after an incident is always who saw it, and that log has to already exist.
- Encrypt media at rest and keep it out of any bucket with public listing. A meeting recording is the highest-value single object in most companies' storage.
- Do not record interviews with this. Hiring recordings are employment records with their own retention, access and candidate-rights obligations, and that is not a weekend project's problem to take on.
- Keep the extension's permissions as narrow as the capture requires, publish what it can see in plain language, and remember that asking a whole team to install it makes you their supply chain.
- Handle a deletion request as a first-class feature with a button, not as a SQL query you run when reminded.
I am building a meeting recorder that captures the call from my own machine rather than
joining as a bot. The pipeline is not the hard part. Work in this order and argue with
me if I try to reorder it.
1. Start with notice and consent, before any capture code: an unmistakable on-screen
recording indicator, an automatic message posted to the meeting chat when capture
starts, and a one-click stop. Tell me plainly that removing the bot removed the only
thing participants would otherwise have seen.
2. Refuse to build a silent or hidden recording mode. If I ask, say why, and which
jurisdictions make it a problem.
3. Default to audio only. If I enable video, warn me in the code and the README that this
captures every screen share, including data belonging to people not in the meeting.
4. Set a retention period before writing the storage layer, implement the deletion job in
the same commit, and enumerate every place a recording lands: media file, transcript,
summary, embeddings, search index, clips, and anything pushed to a CRM.
5. Authorise every read of a recording, transcript, clip and share link against the
requesting user. An unguessable URL is not access control. Write the test that fetches
another user's recording by id and expects a 404.
6. Log reads and shares as well as writes, so "who has seen this" is answerable.
7. Encrypt media at rest, keep the bucket private with no listing, and use short-lived
signed URLs for playback.
8. Keep extension permissions minimal, and list every one and its purpose in the README,
in language a colleague could evaluate.
9. Do not push transcripts to a CRM or ATS. If I insist, say first that this copies the
content into a second system with its own retention and access model, and that
interview recordings are employment records.
10. Mark speaker attribution as unreliable wherever a summary is shown. A confident
misattribution of who committed to what is the failure nobody catches.
11. Out of scope unless I ask: multi-platform capture, team sharing, phone recording.
12. End by telling me what Bluedot costs per seat, and that most of it is consent,
retention and access-control work I would otherwise do at midnight.That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
As soon as anyone outside your own company is on the call. $25 a seat buys consent notices that have been reviewed, a retention setting that actually deletes, per-recording access control, an admin who can answer a data request, and the compliance paperwork your enterprise customers will eventually demand before they let you record them at all. The transcription is the cheap part and always was — the subscription is buying the policy layer around it, which is exactly the layer a weekend build has none of.
$25/mo is cheaper than your weekend.
Keep the media and the transcripts as ordinary files in ordinary formats — MP4 or M4A alongside a plain-text or JSON transcript with speaker labels and timestamps — in a folder named by meeting id, with a manifest of who attended and who was told. If you walk away, that structure is readable by any tool including a human, and it is also the artefact you need if somebody ever asks what you hold about them. The part that does not migrate is the summaries, which are model output tied to a prompt you will have changed six times; treat them as disposable and keep the transcript as the record. And before you abandon it, run the deletion job. An orphaned bucket of meeting recordings is the worst possible thing to leave behind.
Local speech-to-text engine suitable for private transcription with no audio leaving the machine.
Questions
Granola is also YOUR FUNERAL. What is different about Bluedot?
Granola's entry is about consent and a transcript on your laptop. Bluedot adds two things that raise the stakes: it captures video, so screen-shared content belonging to third parties lands in your store, and it is a team product, so recordings become a shared searchable archive with an authorisation model you now have to get right. The consent argument is the same. The blast radius is bigger.
Doesn't 'bot-free' make it less invasive, not more?
It makes it less annoying, which is a genuine product benefit and also the reason to be careful. The bot in the participant list is doing compliance work as a side effect — everyone can see it. Take it away and the notice has to be something you deliberately built, and a notice that only appears in your own extension's popup is not a notice at all.
Where do the recording laws actually bite?
Several US states require every party to consent rather than just one, and which rule applies can depend on where each participant is sitting rather than where you are. In the EU, recording a call needs a lawful basis under GDPR and the participants have rights over the recording afterwards, including access and erasure. Neither of these is exotic — they are the normal case for any call with someone outside your company.
Can I at least build the transcription part?
Yes, and it is the enjoyable part. Whisper running locally over recordings you already have, with no sharing, no cloud storage and a retention job, is a reasonable weekend and keeps the audio on your machine. The line to hold is that it processes recordings you were entitled to make, rather than becoming the thing that makes them.
- Recording phone calls and conversations — consent requirements by state (US)
- GDPR Art. 6 — lawfulness of processing (EU)
- GDPR Art. 5 — principles relating to processing of personal data (EU)
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
A transcript of a call someone didn’t know was recorded isn’t a side project. It’s evidence.
Whisper never says "I couldn't hear that". It writes a plausible sentence instead, and your archive keeps it forever.
Recording the call is the easy half. Being allowed to is the half nobody prototypes.
last reviewed 2026-08-04 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice