Configuring AI providers and defaults

Your finished transcript and speaker labels always run on your own computer, on a model Daisy ships or you download. The Providers section holds both of those things, in two clearly separated blocks: Local transcription (which model transcribes, no key involved) and AI provider — the optional piece that writes your summaries and powers Q&A, chapters, the Analyzer, and transcript polish. Every cloud provider is optional and uses your own API key, billed directly to you. Daisy never proxies or marks up that usage.

You can pick a cloud provider (Groq, OpenAI, or Anthropic), a keyless local server (LM Studio or Ollama), or None to skip AI and use the copy-paste workflow. Transcription always runs on-device.

Add an AI provider

1. Go to Settings and open the Providers section.
2. Under AI provider, pick the provider you want to use (Groq, OpenAI, Anthropic, LM Studio, or Ollama). The section shows a single editor for the provider you select.
3. Select Configure (or Edit if a key already exists).
4. Step 1 of 2 — paste your API key, and adjust the Base URL only if you need to. LM Studio and Ollama are keyless, so there's no key field; leave the base URL blank and Daisy uses that server's usual local address, or fill it in if you moved the server off its default port or onto another machine.
5. Select Test key & list models (or Test connection & list models for keyless local servers). Daisy makes one call to confirm the connection works and pulls the list of available models.
6. Step 2 of 2 — pick a model from the list, or type a custom model name in the field below it, then select Save.

Picking a provider in this dropdown also makes it your default for AI features. To remove a stored key later, reopen the editor and select Clear key.

Local transcription sits above it, in the same section

Providers opens on Local transcription — the on-device model list — with AI provider underneath it. The two are unrelated: no provider, key, or account is involved in transcription at any point, and the section says so at the top: Transcription always runs on-device.

Pick a model there with Use, or Download one you don't have. Full detail in Transcription models and language.

Advanced: transcription tuning

Under Advanced → Transcription tuning you can change how each engine runs on this machine: window, hop, commit margin, encoder threads and CPU spin-wait for Parakeet; language, live window cap and decode threads for Whisper; and the Faint-audio markers toggle. Every group has a Reset to defaults button, and changes apply to the next recording. The defaults are right for most machines — this is here for the ones they aren't.

Set your defaults

In Settings, open the Recordings section to choose what Daisy uses by default. The rows sit in five groups:

Microphone

• Microphone — the input device for new recordings.
• Your speaking level — fine-tune the input level Daisy treats as speech.

Live captions

• Show captions while recording — a per-machine on/off switch, and display only. The transcript is produced either way. See Live and real-time transcription.

Cleanup

• Remove the other side's voice from your mic — Automatic, Always, or Never, for removing speaker echo from your microphone track.
• Reduce background noise — clean up background noise in saved audio. Off by default; tick it to turn it on.

AI features

• Provider — the same provider dropdown as above (summary, Q&A, chapters, analysis, polish). "None" uses copy-paste.

Troubleshooting — these two are per-machine rather than part of your profile:

• Detailed logs — debug logging level.
• Log files — an Open folder button for the logs folder.

Appearance, theme files, and the whole-UI Text size control live in Settings → Display — see Appearance, themes, and text size. Update checking lives in Settings → About — see Staying up to date.