After a recording ends, talat can write a summary, a list of action points, and chapters. All of this is optional, and it involves two settings tabs: LLMs, where you set up the model that does the writing, and Summaries, where you choose that model and what it produces.
Setting up a model
Under Settings → LLMs you keep a list of the models talat can use. Add as many as you like and give each a name; each one is either local or connects to an external provider with your own API key. Your options are:
- Built-in. A model that runs entirely on your machine, with nothing to set up and nothing leaving the box. Good for short meetings and anything sensitive; it tends to get conservative on meetings much over half an hour. If your machine doesn’t have enough memory for it, the option is disabled and the page explains why.
- Ollama. If you already run Ollama, point talat at it and pick any model you’ve pulled. Still fully local.
- OpenAI and Anthropic. Cloud models, better on longer meetings, at the cost of sending the transcript to the provider. Bring your own API key.
- Claude Code and Codex. If you already have a Claude Code or Codex subscription, talat can run summaries and dictation cleanup through the CLI you’ve installed, using that subscription. No separate API key, and nothing to configure beyond picking it.
- Any OpenAI-compatible endpoint. Set the base URL and model name yourself to use another provider or a self-hosted server.
API keys are stored in the operating system credential store (Keychain on Mac, Credential Manager on Windows), never in a settings file. The built-in model can also be set to unload when talat is idle, to free memory, and it reloads on its own when you next need it.
Choosing what talat generates
Under Settings → Summaries, pick which of your models writes the notes, or choose No summaries to turn the whole feature off. With it off, recordings are still transcribed and saved; nothing is generated and no model is loaded.
Once a model is selected, you control:
- Summarise when a recording ends. On by default. You can also summarise a single meeting by hand from its detail view.
- Generate chapters. Topical sections along the playback bar.
- Generate action points. Concrete to-dos the meeting agreed on.
- Open meetings on the Summary tab. When a summary is ready, open straight to it rather than the transcript.
Custom prompt
talat ships with a summarisation prompt that works for most meetings. You can override it on the same tab; your prompt is used for every new summary until you reset it to the default.
For a fuller walk-through of picking the right model, see our guide to summary models.