Meeting notes with your own API key — bring your own LLM

Daisy lets you take meeting notes with your own API key instead of renting a model from a SaaS. The part most tools charge for — recording, the transcript, and speaker labels — happens on your machine and is free, no key needed. The AI layer on top (summaries, Q&A, chapters, coaching) is where a model comes in, and that model is yours to pick: a cloud key you bring, a fully local LLM, or nothing at all. Daisy never proxies the call and never marks it up.

To be precise about the split: the API key is only for the AI features. Recording, transcription, speaker ID / diarization, and full-text search all run on-device and free in every configuration. You could install Daisy, never enter a key, and still walk away with a complete labeled transcript on your disk.

Which model? Your call

There are three honest paths for the AI features, and you can switch between them per meeting. No path is a hard wall — if you skip the model entirely, you still get the transcript plus a prompt you can paste anywhere.

• A cloud key you bring — OpenAI, Anthropic, or Groq, billed directly to you.
• A fully local LLM — Ollama or LM Studio, nothing leaves the machine.
• None — take the transcript and a ready-made copy-paste prompt to whatever chat you already use.

Pick based on what you care about: privacy, cost, model quality, or just what you already have running.

Bring a cloud key (OpenAI, Anthropic, Groq)

Paste a key from your provider and Daisy uses it for summaries, Q&A, chapters, and coaching. The request goes straight from your machine to the provider over HTTPS — Daisy is not in the middle. There's no Daisy account in the loop, no relay server, and no markup on tokens; you pay your provider their list price and nothing else.

Only text goes out: the transcript (or the slice of it a feature needs). The audio never leaves your disk for this. You stay in control of the key the way you always do — set a budget cap in the provider's console, rotate it, or revoke it any time, and Daisy simply stops being able to call the model.

Want a different model for different jobs? Point summaries at one provider and Q&A at another. It's your key and your choice.

Or run it fully local (Ollama, LM Studio)

If you already run models locally, point Daisy at Ollama or LM Studio and the AI features go through that. Now nothing leaves the machine — not the audio, not the transcript, not a single token. This is the same on-device guarantee the core loop already gives you, extended to summaries and Q&A.

The trade is the obvious one: a local model wants the RAM/VRAM to run well, and a small local model won't match a frontier hosted one on a dense hour-long call. But for a lot of meetings, a decent local model is more than enough — and it costs nothing per run.

Or none at all — the copy-paste path

You don't have to wire up a model to get value. Stop the meeting and Daisy hands you the finished transcript plus a ready-made prompt built for it. Drop both into whatever chat you already pay for and get your summary there.

This is the zero-config, zero-extra-cost path. It's also the escape hatch: even if a provider is down or you simply don't trust an integration today, the transcript is yours and the prompt makes it one paste away from notes.

Your keys live in an encrypted vault

A cloud key you paste is stored in an encrypted vault on your own machine — AES-256-GCM, with the key derived from your passphrase via Argon2id. The passphrase never leaves your laptop. There is no reset and no recovery flow, which is the point: nobody at Daisy can read your vault, and nobody can hand it over, because the secret to open it only exists on your hardware.

That means the discipline is yours too. Lose the passphrase and the vault stays sealed; you'd re-enter your keys. In exchange, the keys are genuinely yours — not escrowed on a server, not synced to a cloud you don't control.

What costs what

Here's the qualitative map, no invented numbers:

• Recording, transcript, speaker labels, search — free, on-device, no key, every time.
• A cloud key — you pay your provider directly, typically pennies per meeting at current rates for text summaries. Set a budget cap in the console so there are no surprises; you control and can revoke the key whenever you want.
• A local model — no per-meeting cost at all; you're spending your own compute.
• The copy-paste path — free from Daisy's side; you use whatever chat you already have.

Daisy itself is a desktop app you own, not a per-seat cloud subscription with a token meter. One license covers three devices, verified offline, with no account and no sign-up. Files are saved in open formats — plain-markdown transcripts and standard audio — so nothing is trapped.

It runs on Linux, Windows, and macOS (Apple Silicon .dmg), with a 30-day trial so you can confirm the whole flow — on-device transcript, your own key for the AI — on your own hardware before deciding anything.