talat 1.0.0: new features, new ways to pay

After four months and 37 releases, talat has reached 1.0.0. It's a big milestone for us, and there's a lot in this one:

• Mid-meeting screenshots: use talat’s in-app tool or your computer’s native key combo to add a timestamped screenshot to a meeting. Screenshots are shown alongside the transcript they were captured with, and talat uses on-device OCR to process and index them.
• Focus mode: when you’re mid-call, you mostly only care about what’s being said, by who, and when. Focus mode lets you maximise the transcript view of a meeting, collapsing everything else so you can more easily glance at it to catch up on recent conversation.
• Related meetings: you can quickly jump to any meeting in a calendar series from the meeting page, and you can jump to meetings featuring a given speaker from their profile page.
• Redesigned speakers experience: speakers, and their voices, are central to talat - but the latter used to be hidden away unless you went looking for it. From now on, the concept of a person’s ‘voice print’ is surfaced a lot more clearly as part of a redesigned speakers section, to help you get the most out of talat’s speaker identification.
• Speaker groups: one of the most common transcript-related requests we’ve received over the last few months is “when people from a certain organisation are in a meeting, I want these company-specific words to be recognised”. Speaker groups let you do this in a flexible way: you can associate ‘boosted’ words with a group, and you can add people to any number of groups.

The journey to 1.0.0

Today's 1.0.0 release is the sum of far more than the features listed above: it's the culmination of every release since March. Back then talat was a Mac-only prototype, and the distance from there to here is worth retracing, most recent first.

The latest additions have been about doing more with what you say. talat can now dictate, typing your words straight into whatever you’re working in rather than only transcribing a meeting. Before that came a run of faster, more accurate speech models that pushed talat well beyond English and into dozens of languages. We added file import, so an existing recording gets the same transcript, speakers and summary as a live call. Summaries grew chapters and action items, breaking a long meeting into sections you can navigate and pulling out what everyone agreed to do. And speaker identification got markedly smarter, suggesting who’s talking while the meeting is still running instead of leaving it all until afterwards.

Further back are the foundations the rest of it stands on. Version 0.10.0 redesigned the app from the ground up and taught talat to read your Apple and Google calendars, so a meeting arrives already knowing its title and who’s in it. Before that, talat came to Windows, having started life on the Mac alone. And earlier still, we added an MCP server, so you can point Claude or another assistant at your own meetings and ask questions across them.

That’s slightly over two releases a week since the first public build, and keeping up that pace is a large part of why talat’s pricing is changing too.

New ways to pay

During pre-release you could only buy talat outright at a heavily discounted price. In addition to that lifetime option (now at full price), you can now pay month-to-month, or year-to-year. This avoids the upfront cost, and means you only pay for talat as long as you’re happy with it. There’s no minimum term, and our existing 30-day no-questions-asked refund policy still applies. Together with the full-price lifetime option, it puts talat on a sustainable footing so we can keep improving it.

Updated privacy policy and terms

Graduating to 1.0.0 and adding pay-as-you-go options means updating our terms and privacy policy, and we’d encourage users new and existing to read both.

For pay-as-you-go to work, and to see how talat is doing out in the world, the app now sends a small, fixed ping from time to time: once on launch, then every few hours while it’s running. It never includes anything you record. Not your audio, your transcripts, or your summaries, and nothing about your meetings, your speakers, or their voices; all of that stays on your machine, as it always has. There’s still no account to create, and the ping carries no name, email, or contact details of any kind.

What it does carry is deliberately small: the app version and whether you’re on Mac or Windows, a random per-install identifier that isn’t derived from your hardware or your identity, and how far you got through onboarding. If you’re on a paid licence it adds an opaque, hashed licence id, so we can confirm the subscription is still active and end access cleanly when a plan lapses or is refunded. If you’re on the free trial it reports how much of the trial you’ve used, which is how the trial stays limited without talat ever asking you to sign up.

The road ahead

So, does hitting v1 mean we’re done? Absolutely not. In some ways, talat 1.0.0 is more like our first release - it’s what we would have shipped if we’d squirrelled ourselves away for the last four months instead of shipping and iterating on a pre-release version. Except that’s not quite true, because that version would have lacked the countless bug fixes, improvements and features we’ve added in response to the sheer volume of feedback we’ve had from our early adopters. The feedback we’ve received directly, requests on our roadmap, and suggestions provided by our Discord community have all shaped talat into the product it is today.

Concretely, three near-term improvements coming to talat are:

• Cross-conversation intelligence: today, meetings in talat mostly happen in isolation. Indeed, it’s only with the release of 1.0.0 that we’re adding the first tentative version of related meetings (see the feature list above). But “also in this series” just scratches the surface of connectedness, and we have a lot of work to do to improve this.
• Custom workflows: our hands-down most requested feature is the ability to customise some of the things talat does (or how it does them) based on certain conditions being met. For example, “when I have a one-to-one, use this prompt during summarisation”. We’ve had plenty of requests along these lines, and we want to make sure what we ship covers the cases people actually asked for. I’m really excited to focus on and ship this soon.
• A built-in agent: today, if you want to ask questions about your meetings you point Claude or another assistant at talat’s MCP server. Soon you won’t need anything else: talat will have its own agent, built right into the app. You can type to it, or say “hey talat” while a meeting is running to reach it by voice. It can answer questions about what’s been said so far, or make changes for you like adding a note against the transcript, and it uses the same on-device model as the rest of talat, so nothing leaves your machine.

These are big pieces of work, and there’s plenty more to come; we have no intention of slowing down.

The future is not only about features, either. Underneath everything sit two things we treat as never finished: performance and accuracy. Every release, we work on making talat faster and lighter on your machine, and on getting more of your words right and more of your speakers correctly recognised. We’ve come a long way on both, and there’s a great deal further to go. These axes are what our early users consistently tell us they care about the most, and what sets us apart from the emerging group of local-first transcription apps, because these are hard problems which you can't solve in a weekend.

If you’re already using talat, please keep the feedback coming. If you haven’t tried it yet, download talat and see what a meeting looks like when everything stays on your machine.