Wave

Meeting notetakers have a blind spot they rarely admit to: they assume the meeting is a video call.

A great deal of what is worth capturing is not. The conversation after the meeting ends, the corridor discussion where the real decision happens, the client call taken in a car, the lecture in a hall, the interview across a table.

Wave's answer is to run on everything — including a watch — which sounds like a feature-list detail and is actually the whole product. A recorder that is already on your wrist captures conversations a Zoom bot never hears.

Wave: A Notetaker On Every Device

• It runs on almost everything — iPhone, iPad, Android, web, macOS, Windows, Chrome, Apple Watch and Wear OS.
• 76 languages for transcription and summarisation, with translation between any of them.
• Voice ID identifies speakers automatically, rather than labelling them Speaker 1 and Speaker 2.
• Never trains AI models on customer data, with recordings and transcripts stated as belonging to you.
• A developer API and MCP server, so notes are reachable from Claude, ChatGPT, Zapier and your own code.

Why Does Device Coverage Decide A Notetaker?

Wave runs on iPhone, iPad, Android, the web, macOS and Windows desktops, a Chrome extension, Apple Watch and Wear OS smartwatches, and it integrates with Zoom, Google Meet and Microsoft Teams.

That list is doing more work than a feature comparison suggests. Most notetakers are meeting bots: they join a scheduled call, record it, and produce notes. That covers a specific and increasingly small proportion of the conversations that matter.

The rest happen without a calendar invitation. Someone catches you after a meeting and tells you the actual constraint. A supplier calls your mobile. A customer says something important in a corridor at a conference.

A recorder you can start in two seconds from a watch captures those, and one that requires a laptop, an application and a meeting link does not — because by the time you have opened anything, the moment has passed.

The trade is that ubiquity is worthless without discipline. A tool this easy to start is a tool you will accumulate hundreds of recordings in, and a pile of unreviewed transcripts is the note-taking equivalent of a read-later app.

The established comparison is Otter.ai, which does meeting transcription well and assumes a meeting. Where the two diverge is everything that happens outside one.

Most notetakers are meeting bots. The conversation where the real constraint gets mentioned usually happens after the meeting, without a calendar invitation.

Voice ID And The Speaker Problem

Wave lists Voice ID for automatic speaker identification, which is a step beyond the speaker labelling most transcription offers.

The distinction matters more than it sounds. Diarisation — separating who spoke when — is a solved-ish problem, and it gives you Speaker 1 and Speaker 2. Knowing that Speaker 2 is your client's head of procurement requires recognising the voice.

For a summary, that difference is everything. "Someone said they need it by March" is not actionable; "the finance director said they need it by March" is a fact you can act on and quote.

It also determines whether action items are usable. A list of commitments with no owner attached is a list of things nobody will do, and attaching the right name automatically is the difference between a summary and a record.

Voice recognition also carries an obligation the marketing does not raise. A system building voiceprints of people in your meetings is holding biometric identifiers, and those are treated as a special category of personal data in several jurisdictions. Worth understanding before enabling it across a team.

Never Training On Customer Data Is The Claim That Matters

Wave states that it never trains AI models on customer data, that recordings and transcripts belong to the user, that everything is encrypted in transit and at rest, and that it is SOC 2 Type 1 and GDPR compliant.

The training clause is the one worth reading twice. Meeting recordings are among the most sensitive material a company produces — pricing discussions, personnel matters, unreleased plans, and candid opinions about customers who are not in the room.

A vendor training models on that is not doing anything sinister, and it does mean your commercial conversations become part of a system other people query. A clear commitment not to is the single most valuable sentence a notetaker can publish.

Ownership of recordings and transcripts is the second clause, and it decides what happens when you leave. A tool that owns your archive can make export difficult; one that states the material is yours has, at minimum, made a commitment you can point at.

SOC 2 Type 1 is where the position weakens slightly, and the distinction is worth knowing. Type 1 attests that controls were designed appropriately at a point in time; Type 2 attests that they operated effectively over a period. Type 1 is a genuine step and it is the weaker of the two.

The strictest version of this argument is a clinical one. Nabla Copilot holds SOC 2 Type 2, ISO 27001 and HIPAA because a consultation recording is regulated material, not merely sensitive.

• Get consent before recording — in several jurisdictions it is a legal requirement, not a courtesy.
• Understand that Voice ID means storing voiceprints, which are biometric data.
• Review recordings weekly, or the archive becomes a read-later pile.
• Check whether SOC 2 has moved from Type 1 to Type 2 if that matters to procurement.
• Use the MCP server to query notes rather than re-reading them.

An API And MCP Server Change What Notes Are For

Wave lists a developer API, an MCP server usable from Claude and ChatGPT, Zapier, and exports to Notion, Google Docs, OneNote, Evernote, PDF, Slack and email.

The MCP server is the interesting one, because it inverts the relationship. Ordinarily you open a notetaker to read a note; an MCP server means an assistant you are already talking to can query your entire meeting history without you opening anything.

That turns an archive into a resource. "What did we agree with this supplier about lead times?" becomes a question you ask in the middle of drafting an email, rather than a search you go and perform in another application.

It is also what makes accumulating hundreds of recordings defensible rather than wasteful. An archive nobody reads is dead weight; an archive an assistant can search is a memory, and the difference is entirely about access.

The export list matters for a simpler reason: nothing is trapped. A notetaker whose output only lives in its own interface creates a second place to look, and the tools people actually use — Notion, Docs, Slack — are where notes need to arrive.

Once transcripts exist, interrogating them is a separate job — a grounded notebook such as Gemini Notebook cites back into the source, which a summary does not.

Recording Meetings Without Annoying Anyone

The social problem with a notetaker is larger than the technical one, and getting it wrong costs more than a bad transcript.

The rules below are partly legal and mostly about not being the person who recorded a conversation without saying so.

1. Establish the consent position for your jurisdiction before recording anything — several require all parties to agree.
2. Say you are recording at the start, every time, including for informal conversations.
3. Decide who has access to a recording before you make it, especially for sensitive discussions.
4. Review and act on notes weekly rather than accumulating an archive nobody opens.
5. Delete recordings you no longer need, since a permanent archive of candid conversations is a liability.
6. Test transcription accuracy on your own product names before relying on any summary.

Wave: Platforms And Privacy