What VoxaFlow sends
Your voice and your text stay on your Mac or Windows PC. Here is every connection the app makes, with real example requests.
Last updated: November 8, 2026
In short: VoxaFlow only talks to our own servers, all at Hetzner in Germany: once to download the models, then once a day to check for updates, for the beta and license check and, if you use a pack, for its word lists. No account, no usage data; for the license only a checksum of your hardware ID. Every request names only “VoxaFlow” and the version, not your macOS version or language. Speech recognition and the AI cleanup run on your Mac, so dictation works with Wi-Fi turned off.
The servers the app may contact
The app has a fixed list of hosts (Endpoints.allowedHosts). An automated test fails as soon as the source code or one of the libraries contacts anything else.
| Host | Purpose | Since |
|---|---|---|
models.voxaflow.ai | Speech and language models (mirror, fixed versions) | 0.6.4 |
updates.voxaflow.ai | Update check and downloads | 0.6.4 |
license.voxaflow.ai | Beta and license check; feedback, only when you send it | 0.8 (license), 0.7.5 (feedback) |
packs.voxaflow.ai | Signed word lists of the packages, only if you switch them on; at most one check a day for new versions, without license data | 0.10 |
The website voxaflow.ai is only ever opened in your browser (for example the beta notice), never loaded by the app.
1. Model download
When: once after installing, and when you switch models. About 2.2 to 2.8 GB, depending on the model. A finished model starts without any network request.
The app first asks for the file list, then fetches the files in pieces (so an interrupted download can resume) and checks each file against its size and checksum. A request for the compact language model:
GET /voxaflow/lektor-kompakt-1.7b-2611/resolve/main/model.safetensors
Host: models.voxaflow.ai
Range: bytes=0-4194303
User-Agent: VoxaFlow/1.0.0-rc4
Accept-Language: *
Accept: */*
Accept-Encoding: gzip, deflateThat is the complete request: no cookie, no token, no identifier. Since 0.12.2 this holds for every download. Up to 0.12.1 the open-source library FluidAudio downloaded the speech recognition model (Parakeet, plus the detector that checks whether anyone is speaking) with the macOS defaults, a user agent such asVoxaFlow/13 CFNetwork/3896.100.1.1.1 Darwin/27.0.0 and your preferred languages (Accept-Language: en-US,en;q=0.9). The Darwin version told which macOS version you use. Our network audit found this, and it now also checks the user agent of every request.
2. Update check
When: once a day. You can turn it off under Settings › About. VoxaFlow uses Sparkle, with updates signed by our own key and notarized by Apple.
GET /beta/appcast.xml
Host: updates.voxaflow.ai
User-Agent: VoxaFlow/0.12.3 Sparkle/2.10.0
Accept: application/rss+xml,*/*;q=0.1
Accept-Language: *
Accept-Encoding: gzip, deflate, brNo system profile (Sparkle can send one; it is switched off), no query parameters. Up to 0.12.1 macOS added the preferred languages, since 0.12.2 it is only *. From version 1.0 the feed is /1/appcast.xml. If there is an update and you install it, the app downloads the file named in the feed, a delta of a few MB or the full archive of about 40 MB.
3. Optional: sync via iCloud
When: only if you turn on sync under “iCloud & Data” (from 0.7.0). The app then puts dictionary, snippets, settings, statistics and notes, and optionally the history, into the “VoxaFlow” folder in your iCloud Drive. With Medical or Legal & Tax, professional secrecy mode applies: history and notes then stay on the Mac, unless you turn the mode off because you are not bound by professional secrecy.
This is not a connection of the app: it writes files to a folder on your Mac, and macOS uploads them to your iCloud. A network monitor shows macOS processes such as bird or cloudd for this, not VoxaFlow. Every file is encrypted with AES-256-GCM on your Mac first; this is how a file starts:
VoxaFlow/devices/9DCB2FAD-…/profile.vxenc
56 58 46 31 01 40 25 d2 46 28 1b ec 78 bc 0d f8 VXF1.@%.F(..x...Only your devices have the key: in the keychain and as a recovery code that you enter on further Macs. Neither Apple nor we can read the contents.
4. Since version 0.8: beta and license check
Since 0.8 the app checks the beta license on license.voxaflow.ai (at first launch, then every 7 days) and, if you enter a license key, the bought license (when activating, then every 30 days). The beta request:
POST /v1/beta
Host: license.voxaflow.ai
User-Agent: VoxaFlow/0.12.3
Accept-Language: *
Content-Type: application/json
{"device_hash": "3f9a…", "platform": "mac", "app_version": "0.12.3"}The device hash is a SHA-256 of a fixed prefix and the hardware UUID of your Mac. The UUID itself is never sent, and the hash cannot be turned back into it. It lets one license cover two devices.
When activating, the app also sends the license key; for each check and when deactivating only the license token it received from the server (it contains the license ID, device hash, packages and dates). The answer is a new, signed token, and the app verifies the signature itself. Without internet the license stays valid for 14 days beyond the check date, and dictation always works.
From February 1, 2027 a newly installed app without a license registers the 14-day trial once (POST /v1/trial with device hash, app version and the trial's start), so it runs only once per Mac.
5. Since version 0.10: word lists of the packs
When: only if your license includes a pack and you switched it on: at most one check for new versions a day at launch, and the download of a list when you switch it on. Up to 0.12.3 the app asked for the catalog at launch also without a pack; since 0.12.4 it does not.
GET /v1/index.json
Host: packs.voxaflow.ai
User-Agent: VoxaFlow/0.12.4
Accept-Language: *No license data, no device identifier: the app decides by itself which lists you may use. The lists are signed, and the app checks the signature before using them.
6. Windows beta
VoxaFlow for Windows (a beta since October 17, 2026) has the same fixed list of hosts (VoxaHttp) and sends only VoxaFlow/<version> and Accept-Language: * with every request, no Windows version. Differences from the Mac:
- Models also come only from
models.voxaflow.ai(Parakeet via sherpa-onnx, VoxaFlow Lektor as GGUF), never directly from Hugging Face or GitHub. - Updates: the app checks three minutes after launch and then once a day. It leaves out the query parameters the update library Velopack would otherwise add (architecture, operating system, version):
GET /win/releases.win.json
Host: updates.voxaflow.ai
User-Agent: VoxaFlow/0.1.0
Accept-Language: *- Beta and license use the same request as above, with
"platform": "win". The device hash is a SHA-256 of the same prefix and the Windows machine identifierMachineGuid. - Word lists come with the installer; the Windows app does not contact
packs.voxaflow.ai. - Feedback: the optional diagnostic report contains the Windows version, processor, memory, settings and the log of the last 24 hours. The log contains no dictated text, and your user name is replaced by
%USERNAME%. - No iCloud sync and no learning from corrections: the Windows app does not read text fields of other programs.
Never sent to us
- Your voice, your dictated text, your history and your notes (with iCloud sync only encrypted into your own iCloud)
- What you select or copy, the names of apps, windows or documents
- Dictionary entries and snippets (the same)
- Name, email address, user name, device name, serial number, hardware model
- Usage statistics, word counts, analytics of any kind. The overview in the app is computed only on your Mac. Crash reports only if you send them yourself with your feedback (see below).
- Anything to third parties: no Hugging Face, no cloud AI, no analytics service
“Send Feedback …” in the menu sends your message, the app version and, if you enter it, your email address tolicense.voxaflow.ai. Only with the checkbox a diagnostic report is added: macOS version, Mac model, settings, log entries of the last 24 hours and the latest crash report, without dictated text and without audio. You can read the report first with “View …”. Nothing is sent until you click Send, and no device ID is included. We delete feedback after 24 months.
What our servers keep
- Access logs with shortened IP addresses (IPv4 /16, IPv6 /32), without user agent and languages, deleted after 7 days.
- Anonymous counts: from each update check our statistics derive country, region and city, the app version and a checksum that changes every day, so we can see how many installations are active. The IP address is not stored.
- Servers at Hetzner in Germany, under a data processing agreement. No US cloud services.
The full legal text is in the privacy policy.
Check it yourself
- With a network monitor such as Little Snitch or LuLu you see every connection VoxaFlow makes. Expect
models.voxaflow.aion first launch,updates.voxaflow.aionce a day,license.voxaflow.aifor the beta and license check andpacks.voxaflow.aiif you switched on a pack, nothing else. - Turn off Wi-Fi after the models are downloaded: dictation keeps working.
- An automated test in the app's test suite (
NetworkAuditTests) hooks into every network session of the app and its libraries and fails on any host not on the list above. Our measurement of October 15, 2026 (state of 0.12.2, empty caches, update, license and word list checks due): 165 requests tomodels.voxaflow.ai, one each tolicense.voxaflow.ai,packs.voxaflow.aiandupdates.voxaflow.ai, no other hosts. Every request named only VoxaFlow and the version as user agent, and only*as language.
Found something that does not match this page? Write to hello@voxaflow.ai.