LISTEN2ME - PRIVACY AND DATA USE
Version: 2026-11-03

Listen2Me does not include publisher-operated analytics, advertising, account
registration, remote microphone monitoring, or an automatic upload of recordings,
transcripts or diagnostic reports to Hangry Labs. Only feedback you explicitly
submit is sent to the publisher. This describes the app's behavior, not
a guarantee about Windows, external software, or any server you choose.

LOCAL SPEECH RECOGNITION
With a downloaded local model, microphone audio is processed on this computer
using the selected CPU or supported GPU. Local inference does not require an
inference server or sending recordings to Hangry Labs. Downloads may be required
first. Whisper and Parakeet process completed takes; Moonshine/Nemotron streaming works
while you speak. The microphone is opened only for an explicit dictation action.

HOSTED SPEECH RECOGNITION
If you select a hosted endpoint, Listen2Me sends microphone audio directly to
that endpoint. Compatible streaming sends chunks while you speak, including
before a finish/cancel decision. Already sent audio cannot be recalled. The
server can receive connection metadata and retain or process audio under its own
terms. Requests for discovery, readiness, available models and languages also go
to that endpoint. API credentials, when configured, are sent to that provider.
Choose a provider you trust; use HTTPS outside a trusted local network. Merely
saving the endpoint during installation does not send microphone audio.

OPTIONAL MODEL AND RUNTIME DOWNLOADS
Nothing downloads unless you explicitly select a model/download option. Files
come from the upstream locations in the app's pinned catalog, including
Hugging Face, GitHub, PyPI/pythonhosted.org and Moonshine's download servers.
Those hosts/CDNs see ordinary request information such as your IP address,
requested resource and request headers, and apply their own policies. File sizes
and SHA-256 checks are verified locally. Hardware checks can inspect CPU, RAM,
free disk space and GPU/driver availability locally to choose compatible options.
No microphone audio is needed for those hardware checks.

The Offline Parakeet installer already includes the pinned Parakeet v3 Q8 weights,
CPU runtime and model card. It copies and verifies these local files without a
model download or automatic CUDA download. On a new setup it selects local CPU
transcription; an existing configured installation keeps its speech preferences.
The standard installer and normal application updates do not include model weights.

LOCAL STORAGE AND RETENTION
Settings, model/runtime files and WebView2 browser data are stored in your user
profile, normally under LOCALAPPDATA/Listen2Me. An explicit settings-path override
can change settings and model locations. Settings include your chosen endpoint,
model and shortcut. An existing API-token file stays at the location you configure. A key entered
in Speech setup is saved in a Windows DPAPI-encrypted file under
LOCALAPPDATA/Listen2Me/credentials, bound to the chosen server origin. Settings
contain only its file reference, not the key itself. Windows user/profile access
is required to decrypt it. Choosing a different server does not reuse the old
key automatically. Disabling/replacing a reference can leave encrypted files
until you remove them or the data folder; this is not a secure-erasure claim.
By default, the selected installed local model is loaded when dictation is enabled
and kept ready in RAM/VRAM. This does not open the microphone or send audio.
Memory-saving options load at the start of a take and unload after a chosen idle
period or after each take. Each take starts a separate recognition session. Pause,
model/provider changes and quitting release the worker. No secure buffer-erasure
claim is made.
Recordings and transcript history are kept in process memory, not intentionally
saved as audio/transcript files or sent to the publisher. Up to ten completed
takes can be retained during the session. Clear removes the app's transcript
history, and quitting ends the session. This is not a secure-memory-erasure
promise; Windows paging, crash dumps, external apps and backups are separate.
When text is typed into another app, that app controls its own saving and sharing.

KEYBOARD AND DESTINATION ACCESS
The app observes keyboard pressed states for the shortcut and checks the originally focused Windows window/control
before inserting text. It never returns to previously selected windows or fields.
It does not keep
a general typing history or read destination field contents. It simulates text
input, so text can already be present in the receiving app when you cancel.

INSTALLATION AND REMOVAL
Start with Windows is optional and off by default. Enabling it registers this
app locally for your Windows account and opens it in the system tray at sign-in.
It does not start recording. Shortcut arming and local model loading follow
their separate settings. Disabling it removes the registration, and uninstall
removes the app's startup entry. Windows may independently block startup.

The installer presents the licence and this notice, then optionally configures
speech/downloads using the same verified manager as the app. Setup progress may
show download metadata. Optional installer/operating-system diagnostics can exist
locally. Neither installation nor uninstall sends those diagnostics to Hangry Labs.
If Microsoft WebView2 is missing, the standard installer offers to download its
standalone installer directly from Microsoft and install it after you accept
Microsoft's separate terms. The Offline installer carries the same prerequisite
and does not download it. Both verify the reviewed file's SHA-256 before running
it. Existing WebView2 installations are left alone. WebView2 can collect and send information
to Microsoft, includes Microsoft Defender SmartScreen, and updates automatically
when online under Microsoft's terms and privacy statement: https://aka.ms/privacy
and https://learn.microsoft.com/en-us/microsoft-edge/privacy-whitepaper#smartscreen.
Uninstall preserves settings/models. You may deliberately remove the data folder
and any separately configured token file when you no longer need them.
The installer's Fresh settings option renames existing settings and WebView2
cache into timestamped backup files/folders and starts with default settings.
Downloaded models and external token files are preserved. These local backups
remain until you remove them; they may contain your prior settings/browser data.

OPTIONAL FEEDBACK
The Support Us feedback form can submit the message you write, its feedback type, app version,
Windows platform identifier and an optional reply email to the Hangry Labs website.
No account or reply email is required. Recordings, transcripts, endpoint settings,
token files and diagnostic attachments are not included. Do not put secrets into
your message. The website receives ordinary connection metadata, including your
IP address. The feedback service uses a keyed identifier derived from that IP
address to limit repeated submissions, and controls feedback storage and handling under its own privacy
notice. A successful submission is confirmed by the receiving website; failed
submissions remain in the form during this app session and are not retried
automatically. This is optional communication, not anonymous network transport.

APPLICATION UPDATE CHECKS AND USER-CONTROLLED INSTALLATION
By default, opening Listen2Me checks GitHub's official Listen2Me release feed
once for a newer version. You can disable this startup check in Settings and
check manually using the footer refresh button. No periodic check or automatic
retry loop is scheduled. GitHub/CDNs receive your IP address
and ordinary request metadata. Only choosing Download update saves an installer locally under
LOCALAPPDATA/Listen2Me/updates, checks its size, checksum, publisher signature and
version. Install & restart pauses dictation, starts setup, and closes/reopens the
app. Normal updates preserve settings/models. An installer is never started just
because a release was found. Local update downloads may remain until removed.

THIRD-PARTY SOFTWARE AND WEBSITE
Windows, WebView2, device drivers, hosted ASR providers, download hosts and any
website you visit may have their own network behavior, diagnostics and policies.
This app notice does not describe the Hangry Labs website's hosting logs.
For questions or privacy requests: contact@hangrylabs.app
The app's About links open your browser or email application only when clicked.
These external applications and destinations then apply their own policies.
Startup checks contact GitHub when enabled. Application installers are never
downloaded or installed automatically.
GitHub hosts public installers, documentation and release notes. Visiting those
pages or downloading an installer exposes ordinary request metadata to GitHub
and its delivery providers, separate from local dictation.
Product help: support@hangrylabs.app
Private vulnerability reports: security@hangrylabs.app