Performance and your hardware
Daisy runs entirely on your machine. Recording, transcription, and speaker labels never leave your computer and are never offloaded to a server. That's the point of Daisy — and it also means your hardware sets the speed. There's no cloud to hide a slow machine behind.
If you want your finished transcript in seconds, use an Apple Silicon Mac or a desktop with a GPU. On a laptop, expect a few minutes per meeting. Here's the full picture so nothing surprises you.
What's fast on any machine
Two things keep up almost everywhere:
- Live transcription while you record. Even a modest laptop transcribes several times faster than people actually speak, so the running transcript stays with the conversation.
- Speaker labeling (diarization). Done on-device, and it's not the slow part.
So during the meeting itself, hardware rarely gets in the way.
Where hardware shows: finalizing the transcript
When you stop recording, Daisy makes the full, clean transcript. This is the step that leans on your machine, and the gap between a fast and a slow computer is large. Rough expectations for a one-hour meeting:
| Your machine | Finished transcript |
|---|---|
| Apple Silicon Mac | Well under a minute |
| Desktop with a GPU | Around a minute |
| Recent laptop with a working GPU | A few minutes |
| Older laptop, no usable GPU | Several minutes |
Daisy uses your GPU automatically where it can — Metal on Mac, Vulkan on Windows and Linux — and falls back to your CPU when no GPU is available. The CPU path works, but it's roughly three times slower than the same machine's GPU. A four-year-old ultrabook with no dedicated graphics is the case where finalizing takes the longest.
There is no cloud transcription option to speed this up. That's a deliberate trade: your audio never leaves your machine, and in exchange a slow machine stays slow. If sub-minute turnaround matters to you, it's a hardware choice, not a setting.
The summary is a separate step
Summaries, chapters, and analysis are not transcription — they need an AI model, and that's often the slowest part on a modest machine:
- A local AI model (LM Studio or Ollama) can take a couple of minutes for an hour-long meeting on a laptop.
- A cloud key (Anthropic, OpenAI, or Groq) turns that into seconds — only the transcript text is sent, never your audio, billed to your own key. See Configuring AI providers.
- No AI at all — use the copy-paste workflow and summarize in your own ChatGPT or Claude. See Running Daisy fully offline.
A cloud key speeds up summaries only. It never changes how long transcription takes.
If your machine feels slow
- Confirm the GPU is being used. It's automatic, but a machine with no graphics driver falls back to the slower CPU path. Updating your GPU driver is the single biggest speed-up on a PC.
- Keep the default transcription model. The bundled
base.enis the fastest and already close to the large model on clean meeting audio. Only switch tolarge-v3-turboif you have a GPU. See Transcription language. - Live captions off is normal on a laptop. Daisy runs a one-time speed check per machine; if your laptop doesn't pass, live captions default off to save battery and keep the fan quiet. It has no effect on the finished transcript. See Live and real-time transcription.
- Bring a cloud key for fast summaries if the AI step is what feels slow, rather than the transcript itself.