Yak records your meetings, transcribes them and writes up what was decided. You can see every service that touches your data, and you choose which model does the writing.
Two things happen to a recorded meeting. It gets transcribed, and then a language model turns that transcript into a summary. Both steps involve a third party, ours included, and we say so plainly. What we couldn't find was a tool that would tell us who those third parties were, or let us swap them for ones that fit our policy.
That matters more than it sounds. Approved AI vendors differ from one organisation to the next. Some can't use Google, some can't use Anthropic, some can only use what already sits inside their existing agreements. A tool with one hard-wired model either fits your policy or it doesn't, and you usually can't tell which until you've asked questions nobody wants to answer.
So we built Yak with the path visible and the summary model chosen by you. Every meeting at Publica has run through it since July. It is the tool we use ourselves, built by our own developers alongside the people using it every day.
Start a recording from the Mac app or your phone. Yak works out which calendar invite it belongs to, so you're not naming files afterwards.
You get the full transcript alongside a written summary covering what was discussed, what was decided and who is doing what next.
A meeting belongs to the person who recorded it. Nobody else in your organisation sees it unless they are given access.
Record a short sample for each person once and Yak knows who is who from then on. Meetings come back with names against the words rather than Speaker A and Speaker B.
Yak recognises the meeting from your calendar and writes it up in the right format. Your sales meetings always come back as your sales format. A morning huddle comes back as actions and follow-ups, and nothing else.
Yak connects directly to Claude and other AI tools, so a meeting can be pulled into a chat or a project without anyone copying and pasting it there. Ask a question and get an answer drawn from everything you have access to.
The model that writes your summary is a setting, not something hard-wired. We run Gemini today because it fits our own policy. If yours points at Anthropic, OpenAI or something already inside your Microsoft agreement, we connect that instead.
Single sign-on, keyed API access between every service, and no shared visibility between users. Nobody sees a meeting that isn't theirs.
Most organisations are fine with the hosted service. For the ones that are not, we build it somewhere they own instead.
You use Yak as a hosted service and we look after the infrastructure, the updates and the support. It's the quickest way to start and the cheapest way to run it.
Yak runs inside your environment, on your infrastructure, in the region you choose, using the transcription and language models you approve. You set retention, access and audit rules, and no Publica staff can see your data.
Yak under your own brand, for your workforce or your clients. We build and maintain it, you own the product and the relationship.
This is usually the first thing a risk team asks for, so we have put it on the website rather than making you ask for it.
Audio is captured on your Mac or phone, and nothing leaves the device while the meeting is still running.
When you stop recording, the file is uploaded to the chosen environment. On a subscription this is Yak. On a private deployment it goes to your own hosting partner.
The file is passed to your transcription model of choice, and then deleted. How quickly varies by model. With our default, Google, it is deleted immediately.
The transcript text is sent by API to the summary model you have chosen, which reads it, writes the summary and returns it. Nothing is retained at that end, and no vendor uses it for training.
The audio file sits in the chosen environment and the transcript sits in the database, either Yak's or your own. On a private deployment both of those are inside your environment, configured at setup, and nothing is held by us.
Transcription runs on Google by default. Other providers can be connected where a policy or an accuracy requirement calls for it.
Here is how Yak handles your information, in plain terms.
Two external services touch a meeting: the transcription provider and the language model that writes the summary. Both are named above, and we will send the full flow as a document for your risk team on request.
The summary model is selected per organisation, so it can be set to whatever your AI policy already permits. Transcription can be changed too, though that is a configuration we do with you rather than a setting you flip.
Our current list is published and customers are told before it changes. On a private deployment you choose the processors yourself.
No customer audio, transcript or summary is used to train any model, ours or a vendor's. That sits in the contract rather than in a promise on a website.
Retention is configurable, and audio, transcripts and summaries can each be treated differently. Audio can be deleted the moment the transcript comes back if you would rather not keep it at all.
On the hosted service a meeting is visible to the person who recorded it and anyone they share it with. A small number of our engineers can reach data for support purposes and that access is logged. On a private deployment we have no access at all.
You can export everything or delete everything on request.
We'd love to have a Yak with you and show you how it works.