AI Note-takers That Record In-Person Meetings

Granola is a desktop application rather than a meeting participant. You open it, press start, and type whatever you would have typed anyway; it listens to your microphone and your computer's audio at the same time and keeps a transcript alongside your notes. When the call ends it writes a summary from both, which is a different product from one that summarises a recording you never looked at.

The trade-off is stated plainly by the vendor: it will not join or record anything on its own. That costs you the meetings you forget to start it for, and buys you a tool that never appears in someone else's participant list. For a lot of people that is the whole argument, and it is why the free plan is worth taking seriously — it carries chat across your past meetings alongside custom note templates, a pairing most vendors hold back.

The cost is real, and it is about data rather than features. Everything is stored on AWS in the United States with no regional option, notes and transcripts are kept indefinitely unless an administrator configures otherwise, and anonymised data is used for training until you turn that off in settings. Audio itself is deleted once transcription finishes.

Fireflies sends a recorder into the meeting and gets out of the way. It can be invited to a single call or set to join everything on your calendar, and once it is running the output lands wherever you have pointed it — a chat channel, a CRM record, or another system through the API.

What makes it worth a close look is where the paywall sits. Unlimited transcription, unlimited summaries, live notes, meeting search, the assistant that answers questions about past calls, and API access are all on the free plan. The cap is storage rather than usage: four hundred minutes of recordings for the whole team, after which older meetings have to be deleted to make room. Paid tiers raise that and add integrations, conversation analytics and administrative controls.

The thing to weigh is what happens to the data afterwards. The vendor states that meeting data never feeds AI model training and describes a zero-day retention arrangement with its own vendors. It holds SOC 2 Type II and GDPR, and HIPAA with a business associate agreement on the enterprise tier. Custom data retention is an advanced-tier control rather than a default.

Read AI produces a meeting report rather than a set of notes. Alongside the transcript and summary it measures engagement and sentiment, breaks the dialogue out by speaker, and shows an in-meeting dashboard while the call is still running. Whether that is insight or noise depends entirely on what you are trying to do with it.

Coverage is a strength. Twenty-six languages are enumerated for meeting reports, and in-person meetings, live events and mobile capture are all documented rather than implied. The one documented limit is mixed-language meetings, which the vendor says are not fully supported and may produce inconsistent results.

The thing to weigh is location. On privacy it sits in the middle, honestly documented. Data is stored encrypted in an Amazon data centre in Northern Virginia, with other locations available only by arrangement through the sales team. Retention is genuinely in the user's hands: you choose whether audio and video are saved for playback, and if you decline, nothing beyond what is needed to produce the report is retained. Premium integrations covering Salesforce, HubSpot and webhooks sit on a paid tier.

Circleback is unusually restrained. The marketing site is nearly bare and the product follows suit: what you get after a meeting is a short set of decisions and action items rather than a long document to skim. Search and questions run across every conversation you have had, and automations can fire once a meeting ends.

Its strongest suit is disclosure. The company is SOC 2 Type II and EU-U.S. Data Privacy Framework certified, commits that customer content never feeds model training, gives workspace administrators control over how long meeting data is kept, and publishes its security portal openly, including a report from an external penetration test — where a request form is the more common arrangement.

Coverage is broad, with over a hundred languages advertised and the list actually published, plus in-person recording on the individual plan. The cost is the entry price: there is no free plan to sit on, and Slack support arrives on the team tier. For a small team that wants short notes and a straight answer about security, that may be a fair trade.

Fellow treats a meeting as something with a beginning rather than only a recording. Agendas, notes and follow-ups sit in one place, and the AI recap is attached to a structure rather than produced in isolation. For recurring meetings — one-to-ones, team syncs, anything with continuity — that changes what the notes are for.

Templates are a real part of the product: purpose-built AI note templates for one-to-ones, sales calls, project syncs, interviews and all-hands, with custom templates you can author yourself. Organisation-wide templates arrive on the business tier, along with multi-calendar sync on enterprise.

The privacy position is stated plainly: Fellow never trains its AI on customer data, and offers granular recording policies for controlling what is captured and when. The cost is a gap in disclosure: no transcription language list exists publicly, leaving coverage unrecorded and the entry outside the language filters. For an English-speaking team that is immaterial; for anyone else it is worth asking before committing.

Jamie's proposition is narrow and clearly stated: no bot joins your call, storage and processing are resident in the European Union, and customer content stays out of model training. For a European team with a procurement checklist, that combination answers most of the questions before they are asked.

Because it captures device audio rather than joining as a participant, it works for in-person conversations as well as online ones, and it does not depend on which meeting platform your counterpart happens to use. Meeting audio is deleted once transcription finishes, so what persists is the text.

The cost is documentation rather than capability. The documentation is the weak spot. Language coverage is advertised as over a hundred for notes and ninety-nine or more for summaries — two different figures on the same page — and no itemised list exists on any page that could be read, which leaves the filters empty for it. The documentation site refused three separate fetch attempts. Salesforce, HubSpot and similar connections arrive on the Pro tier, alongside custom and shared templates.

Bluedot records without putting a participant in your meeting, and leads with privacy in its own marketing. That combination is becoming less rare, but the way it draws its pricing tiers still is.

Bluedot draws its tiers the other way round from the usual arrangement. Slack, Notion, Zapier and Make connections are on the free plan, while custom note templates wait until Pro and CRM and applicant-tracking integrations until Business. If your bottleneck is getting notes in front of a team rather than shaping how they read, that is the more useful way round.

Webhook access starts on the Basic tier, with organisation-wide webhooks and MCP on Business. The catch is silence on languages: Bluedot names no transcription list anywhere public, so its coverage is unrecorded here and the language filters pass it by. For a team working in English with a Slack habit and a dislike of meeting bots, it is a comfortable fit; anyone who needs a specific language should ask first.

Notion's meeting notes are worth separating from the workspace they live in. The capture model is closer to a dedicated desktop notetaker than to a suite feature: the app records the system audio and your microphone, so nothing joins the call and the other participants are not notified by a new attendee appearing.

The browser version captures only the microphone, which the documentation suggests suits in-person conversations, while the desktop app handles both sides of a video call. Either way the result is a Notion page, which is the real argument for it — if your team's notes already live in Notion, this removes an export step and a second subscription.

Customisation is genuinely open: the summary template is an ordinary Notion page you edit to define your preferred structure and instructions. The cost is commercial and informational: it requires a Business or Enterprise plan, and the languages it can handle are documented nowhere at all.

Krisp started as noise cancellation and the note-taking grew out of it. That heritage is the reason to consider it: the audio layer is the mature part of the product, with real-time noise removal, accent conversion and voice translation built for contact-centre work before anyone was writing meeting summaries.

Capture happens at the device, so nothing joins your call, and in-person meetings and outbound phone calls are covered as well as video conferences. If your problem is a noisy room or a hard-to-follow accent as much as it is note-taking, few note-takers approach the problem from that direction.

The cost is scope: as a notetaker it is narrow. The help centre enumerates sixteen transcription languages — a list without Japanese, Mandarin, Turkish or Arabic — while the product page advertises seventeen or more, and this directory scores from the enumerated list. There is no free plan to settle into, only a seven-day trial of the premium features, and the summarising and workflow features are thinner than the specialists offer.