The first time you launch talat it opens on a setup sheet, the same on Mac and Windows. The six steps are listed down the left, and the right of the window shows a preview of what your current choice does. Only the first step, granting permissions, has to be completed to continue; the rest have sensible defaults, and every choice can be changed later in Settings.
A step you’ve answered closes to a single row showing your answer. Click a row to reopen it and change the answer. From the second step on, Skip setup opens the app straight away. Answers you have already given are kept; if you skip before answering the Summaries step, automatic summaries are left off.
1. Permissions
Grant access to your microphone and system audio. See the permissions page for the per-OS details.
2. Language
Tell talat whether your meetings are in English only or a mix of languages. This chooses the on-device speech model it uses: an English-tuned model that’s a little faster and sharper, or a multilingual model that also handles languages like Japanese, Arabic, and Korean. On Windows, talat checks what your machine can handle and offers a lighter English model to the ones that need it. You can change this later in Settings → Transcription.
talat downloads the model you pick in the background, so there’s no separate download screen to wait on. If it hasn’t finished by the time you reach the main app, the status line at the bottom of the window shows progress, and a recording you start early begins the moment the model is ready. Windows has a little more to fetch than Mac, so give it a moment on a first run.
3. Recording
Choose when talat starts recording:
- Automatically. talat watches for video-call apps and starts recording on its own.
- Ask first. talat notices the call and asks if you want to record it.
- Only when I say so. Nothing records until you start a recording yourself.
Your choice applies to every meeting app. You can set each app on its own later under Settings → Meeting apps.
4. Summaries
Choose whether talat writes a summary, action points, and a title at the end of each meeting. Pick Yes, on this computer to use the small local model that runs on your machine, with nothing to set up. Pick Not automatically to leave it off; you can still generate a summary by hand for any meeting, and turn automatic summaries on whenever you like.
Later, in Settings → LLMs, you can point summaries at a cloud provider (OpenAI, Anthropic, or any OpenAI-compatible endpoint) or a local Ollama instance, using your own API key. Cloud providers tend to produce better summaries on longer meetings; the transcript is sent to the provider you choose, and only then. If your machine doesn’t have enough memory for the local model, this step says so and points you to the other models you can set up later.
If you use Claude Code, Codex or another MCP client, you can connect it to your meetings later under Settings → MCP.
5. Your name
Tell talat your name. This creates your own speaker profile, so your lines are labelled as you across every meeting, and talat can assume the microphone is you when that makes sense. It’s optional, and you can set or change it later from the Speakers area.
6. Speakers
This step has nothing to choose. It explains how talat learns who is speaking: a voice it hasn’t heard before appears in the transcript as Unassigned, and you click it to name the person. talat may then save a voice print and recognise them in later meetings. Naming people early, and correcting mistakes as you spot them, gives the best results.
Finishing
Once all six steps are answered the sheet shows every choice as a row, and any of them can still be reopened and changed. A Launch at login switch appears below the rows; it is off unless you turn it on. Get started opens the app in the same window.
Until you have recorded a meeting, the Meetings page shows a few other things talat can do: dictating into any app, connecting a calendar, importing recordings you already have, and connecting an AI assistant over MCP.