Questions, answered short.

Whozit turns a meeting recording into Markdown with the right person's name on every line. Most transcripts label voices Speaker 1, Speaker 2 and so on, which is hard to use. With Whozit you name each person once, and later meetings name the people it recognises by themselves. Claude Code, Cursor and other MCP clients can then read your meetings with the names in place.

After a recording is transcribed, Whozit plays each voice so you can listen, then you click a name. Usually that is the only time you name that person: in later recordings Whozit names the people it recognises by itself. A voice it isn't sure about is left for you, never guessed.

No, it doesn't guess. When it isn't certain, it marks the line instead of picking a name.

• In the app, a blue ? is a voice Whozit isn't sure of.
• In the Markdown, an unnamed voice shows as Speaker 4 (might be Kai Tanaka). The name is only a hint.
• A name nobody has checked shows as Kai Tanaka (name unconfirmed). Whozit filled it in from a voice it knows.

A hint is never written as a name. Confirmed names are the ones a person put there or checked, and agents that read your meetings are told which is which.

Yes. Upload a file from any phone, recorder or app: m4a, mp3, wav, mp4, aac, ogg, opus, webm, flac, mov, mkv, wma, amr, 3gp, caf or aiff. A recording saved from a meeting tool like Zoom works too, if it is one of those formats. Video is converted to audio. Each file can be up to 500 MB.

The names don't come from the file or the app that made it. You name each voice once in Whozit.

Yes. One phone in the middle of the room works. Whozit tells the voices apart and you name each one. But overlapping speech and short interjections can land under the wrong voice. That is why lines Whozit is unsure about are marked instead of guessed.

No. You upload the recording yourself, after the meeting. Nothing joins, listens in or records your calls.

Whozit has been tested on English meetings. We haven't measured other languages, so we can't promise how well they work.

Your recordings, transcripts and voices are kept in a space of your own on Whozit's server, separate from other users'. An MCP token you create only reads your data.

To transcribe a recording, Whozit sends the audio to ElevenLabs Scribe. Straight after, Whozit asks ElevenLabs to delete its copy, and retries until it is gone. Any AI tool you connect sees what it reads from your meetings. The privacy page has the details.

Whozit keeps a voiceprint, a description of how a person sounds, for each person you name, so it can name them in later recordings. Before it makes voiceprints from your own recordings, you agree to a consent statement, and Whozit keeps a dated record of it.

Whether that is legal depends on where you and the speakers are. Some places require everyone's consent to record, and some regulate voiceprints. Getting that consent where the law needs it is up to you, and if you're unsure, ask a lawyer. The terms say more.

You can delete it. Forget a person and Whozit drops what it learned about them, or delete everything or your account.

Yes, all of it, at any time. You can export everything as Markdown (one file per meeting, or a zip), forget a person, delete everything you've uploaded, or delete your account.

Whozit has a hosted MCP server. Today it works with any MCP client that can send a header, including Claude Code, Claude Desktop (through mcp-remote), Cursor and VS Code. You make a personal token on your account page and paste it in. The token only reads your data.

The Claude.ai and ChatGPT connector directories need OAuth sign-in, and Whozit doesn't have that yet, so it isn't in them. Setup for each client.

The first 5 hours of recordings are free, with no card. That is 5 hours in total per account, not per month, and it is recording time: a 45-minute meeting uses 45 minutes. Deleting a recording doesn't give hours back. The free hours get you the same naming, Markdown export and MCP access as the plan.

The plan is $15 a month, or $144 a year ($12 a month), with 20 hours of recordings a month. Paid plans aren't open yet, so today the free hours are what you get. See pricing.

• No bot joins your calls. You upload a recording afterwards, from any device. A phone on the table works.
• You name each person once, and later meetings come back named.
• A voice Whozit isn't sure of is marked “?” instead of guessed. Many notetakers label everyone in a room with one phone as Speaker 1, Speaker 2.
• The Markdown is yours to keep, and Claude Code and other MCP clients can read it.

Yes. The built-in sample team has four people and three short meetings from the AMI Meeting Corpus (CC BY 4.0). Try the sample team to name people and see the Markdown.