Local meeting transcription that never leaves your computer

Daisy does local meeting transcription the way you'd actually want it: the audio is captured on your machine, turned into a transcript on your machine, and saved to your disk as files you own. No upload step. No bucket somewhere with your standups in it. The recording never has to leave the computer it was made on.

That covers the whole core loop — recording, the transcript, speaker labels, and full-text search — and all of it is free. The cloud only ever enters the picture for two narrow, opt-in things, spelled out below so you can decide before you install.

What "local" actually means here

Local means the work happens on your CPU/GPU and the output lands on your filesystem. Daisy has no account, no sign-up, and no phone-home telemetry. Your license is verified offline; one license covers three devices.

The files are open and yours: raw audio as .wav, a compact stereo .opus for keeping, and the transcript as plain markdown. You can grep them, back them up, feed them to whatever tool you like, or delete them. Nothing is locked behind a proprietary blob or a server you don't control.

How Daisy transcribes on-device

Recording, transcription, speaker identification (diarization), and search all run locally and free.

Worth calling out: diarization is always on-device and free, no matter how you transcribe. Speaker labels never depend on a cloud key — that work stays local in every configuration. If you've used hosted tools where "who said what" is a paid API call, this is the differentiator.

When you stop a meeting, Daisy finalizes the full transcript locally on normal hardware. A one-hour meeting finishes in a few minutes on a typical machine — no GPU farm, no queue, no upload. That finalized transcript is the canonical artifact; everything else is layered on top of it.

Where the cloud is optional (and where it isn't)

Two things sit at the edge. Both are opt-in, and the core loop above never touches them.

Live captions — words appearing on screen as people talk — are the hardware-sensitive part, so think of it as a ladder rather than a single switch:

• Desktops and Apple Silicon Macs run on-device live captions directly.
• Lighter laptops choose: run live captions on-device (heavier on the machine) or bring your own Deepgram realtime key for a lighter footprint.
• Or skip live captions entirely and just take the full transcript when you stop — same finished result, no live overlay.

That BYOK Deepgram key is the only place the cloud appears in transcription, and only for live captions on lighter laptops. Deepgram is never the general transcriber; the finalized transcript is always local.

Summaries, Q&A, and coaching need a language model, which is the one genuinely non-local step — and only if you don't run your own. You have three options:

• A local LLM via Ollama or LM Studio — fully private, nothing leaves the machine.
• A cloud key you bring (OpenAI, Anthropic, Groq) if you'd rather use a hosted model.
• None at all — Daisy hands you the transcript plus a ready-made copy-paste prompt, so you can run it wherever you like.

That's the honest map. Private by default, your choice to extend.

What you give up — honestly

No magic, so here's the trade:

• Live captions on a thin laptop cost something. Either CPU heat doing it on-device, or a Deepgram key (and a network round-trip) to keep it light. The fully-local-and-instant-captions combo really does want a desktop or Apple Silicon.
• Summaries aren't free physics. Good ones want a capable model. A local LLM keeps it private but needs the RAM/VRAM to run well; a cloud key is easier but sends transcript text out.
• No hosted convenience layer. There's no web app to open your meetings from your phone, no shared team workspace in the cloud. The files are on your disk by design — syncing or sharing is on you.
• It's a desktop app you run, not a bot that joins calls. You record the audio in the room or on your machine; nothing auto-dials into a meeting for you.

If those trades read as features rather than costs, you're the audience.

Who this is for

People who want the meeting record to stay where the meeting happened. Privacy and local-first folks who'd rather run a local LLM than hand a transcript to a SaaS. Anyone under a policy — legal, clinical, regulated, or just personal — that says recordings don't go to someone else's cloud.

On price: free tools tend to pay for themselves with your data. Daisy charges a modest fee instead, so your meetings stay yours.

Daisy runs on Linux (AppImage and .deb), Windows (.zip or NSIS installer), and macOS (Apple Silicon .dmg). There's a 30-day trial with no account and no email, so you can confirm the local loop on your own hardware before deciding anything.