What actually happens to your voice.
- Flume records only when you tell it to — while you hold or toggle the dictation key, or while a meeting you started is running. There is no always-on listening.
- Your audio goes to a speech provider to become text, then the text is tidied by a language model. Both calls route through our own server, which holds the provider keys — your device never does.
- Signed in, your dictations, notes and recordings sync to your account so your other devices see them. Recordings sit in private storage only you can open.
- We don't sell your data, run ads, or train models on your content. Meeting text is never sent to analytics.
- You can delete all of it from inside the app — Settings → Delete account wipes every row, every recording, and the account itself.
Who this covers.
Flume is a voice-dictation app for macOS, Windows, iOS and Android, built by IDIAZ (idiaz.io). This policy covers the Flume apps, this website, and the backend they share. "We" means IDIAZ; "you" means the person using Flume.
Flume is in public beta. When the behaviour described here changes, this page changes with it and the date moves.
Signing in.
Flume accounts are created with Google Sign-In through Supabase Auth. We receive your email address, your name and your profile picture. We never see your Google password.
Your account is what links your Mac, PC and phone together — it's the reason your dictionary follows you between them.
What Flume collects.
Audio you record
When you start a dictation, Flume captures microphone audio for as long as you hold or toggle the dictation key. It's downsampled to 16 kHz mono and sent to a speech-to-text provider to be transcribed. When you record a meeting on desktop, Flume also captures your computer's system audio — the other participants — mixed with your microphone, for the length of that meeting only.
Recording is deliberate and visible: a floating indicator sits on screen the entire time one is running. Flume does not record between dictations, and it never starts a meeting capture without you confirming it.
The text that comes back
The transcript, and its cleaned-up version, are the product. Signed in, they're stored in your account so your devices stay in sync. Alongside each dictation we keep the time, the duration, the device, and the name of the app you dictated into — that's what powers Insights. We never store the contents of that app or anything else on your screen.
Things you create in Flume
Notes, your custom dictionary and snippets, canvas items (text, links and images you send between your devices), and meetings — their transcripts, your scratchpad and the AI summary.
Diagnostics and support
We keep a minimal usage ledger — a timestamp, the kind of request, and an identifier for who made it — for rate limiting and capacity planning. It never contains your audio or your text. If you use Report an issue, we receive your message, your email if you're signed in, your platform, app version, OS version, device name, and a screenshot only if you deliberately attach one.
Where your audio and text go.
Flume's own server is the only thing holding our providers' API keys — they're never shipped inside the app. Every provider below is a processor acting on our instructions.
| Provider | What it receives | Why |
|---|---|---|
| Groq | Your recorded audio; your transcript | Speech-to-text, and the AI cleanup that strips filler and fixes punctuation |
| Google (Gemini) | Your recorded audio | Fallback transcription when Groq is unavailable |
| Ollama Cloud | Your transcript | Fallback for text cleanup and meeting summaries |
| AssemblyAI | Meeting audio only | Working out who spoke when, so a transcript can be labelled by speaker |
| Supabase | Everything stored in your account | Database, file storage and sign-in |
| Resend | Your email address | Account, invite and support email |
| Stripe | Your email and payment details | Flume Pro checkout, invoices, and subscription status. Stripe never receives your audio or transcripts. |
We do not sell your data, we do not share it for advertising, and we do not use your recordings or transcripts to train models. Our providers process your content to return a result to us; what each does with it is governed by their own terms, linked at the foot of this page.
If the cloud providers can't be reached, the desktop app falls back to a Whisper model bundled inside Flume and transcribes locally. That audio is never uploaded.
A little context travels with your transcript
So it spells your terms correctly, Flume sends the words from your own dictionary and the name of the app you're dictating into along with the cleanup request. You can switch this off in Settings.
What stays local.
Flume keeps a working copy of your settings and recent activity in ~/.verbal on macOS and %USERPROFILE%\.verbal on Windows — configuration, dictionary, recent dictations, daily word count, local copies of recordings, and logs. On mobile the equivalent lives in the app's private storage.
Deleting the app removes it; on desktop you can delete that folder yourself whenever you like.
What a team can see.
Join a team and its shared dictionary and snippets are visible to every member. Your dictation counts, durations and the names of apps you dictate into can be shown to your team — but only if you turn sharing on, and it's off until you do.
Your transcripts, notes, recordings and meetings are never visible to your team, your admins, or the team owner.
Accessibility, and what we do with it.
Flume asks for macOS Accessibility (or Windows UI Automation) for one core reason: to put your text where your cursor is. Two optional features use the same permission, and both are worth understanding.
- Auto-learn (off by default) briefly watches the field it just pasted into, so if you correct a misheard word it can offer to remember the fix. The comparison happens entirely on your computer, it stops after 30 seconds, and it skips password fields and terminals. Nothing it reads is uploaded — only the replacement rule you approve.
- File tagging reads the names of open files from a supported code editor, so saying a filename inserts a real reference. Only names, never file contents; they're stored locally and used as a hint to the transcription model.
Both can be switched off in Settings, and Flume works fine without either.
Recording other people.
Meeting capture records both sides of a conversation. Recording other people may require their consent where you live — that's on you. Flume never joins a call as a bot and never announces itself, so nobody else will know a recording is running unless you tell them.
Meeting audio is stored privately in your account. You choose how long it's kept — 7, 30 or 90 days, or indefinitely, which is the current default — in Settings. Meeting transcripts and summaries are never sent to analytics.
How long we keep it, and how to erase it.
Your content stays until you delete it. Individual dictations, notes, meetings and canvas items can be deleted from inside the app at any time.
To remove everything, use Settings → Delete account. It deletes your dictations, notes, dictionary, canvas, devices, meetings and usage rows, deletes your stored recordings and meeting audio, then deletes the account itself. It can't be undone and we can't recover it afterwards. You can also email us and we'll do it for you.
Support reports and the usage ledger are kept separately from your account, for troubleshooting and abuse prevention.
What you can ask for.
You can see your data in the app, correct it, export your notes as Markdown or plain text, and delete any or all of it. Depending on where you live you may also have the right to a copy of your data or to object to certain processing — email us and we'll sort it out.
You can withdraw consent for microphone access, accessibility features or team sharing at any time, in your OS settings or in Flume's.
How it's protected.
Data is encrypted in transit. Recordings and meeting audio live in private storage, opened only through short-lived signed links generated the moment you press play; database access is restricted per account by row-level security. Provider API keys exist only on our server.
No system is perfect. If you find a security problem in Flume, please email us rather than posting it — we'll respond.
Age.
Flume isn't designed for children. Don't use it if you're under 13, or under 16 in the EEA and the UK. We don't knowingly collect data from children; if we learn we have, we delete it.
Where processing happens.
Flume's backend and its providers are in the United States. If you use Flume from elsewhere, your data is transferred there and handled under this policy.
Ask us anything.
Questions, requests, or a deletion you'd rather we handled: szaidi@idiaz.io. A person reads it.
If we change how Flume handles your data, we'll update this page and move the date at the top. Significant changes will be announced in the app or by email.