What leaves your machine: almost nothing.
Terera reads your scripts and hears your voice, which is why speech recognition runs on your computer. This page explains its model downloads and licence checks, and how to inspect its network behaviour.
The complete network behaviour
The app connects for two purposes: getting its speech model and managing your licence. Setup and validation can involve more than one request.
| When | What | What's in it |
|---|---|---|
| Model setup or repair | About 190 MB downloads from the model's public repository on Hugging Face. The installed model runs locally. | The download host can see standard connection information, including your IP address. Your script and microphone audio are not part of the download request. |
| Activation and periodic licence checks | Keygen checks the licence and registers each machine against one of your three seats. A failed network connection alone does not lock you out. | Your licence key and a device identifier are used for validation. Your script, audio and rehearsal history are not included. The privacy policy explains the licence and purchase information held by our providers. |
Speech recognition runs on your processor and audio is discarded as it is transcribed. Your scripts, formatting and rehearsal history stay in your Windows profile. You activate with a licence key rather than creating a user account. New versions are available from the download page.
Verify it yourself
Don't take the table's word for it. Terera behaves identically under observation:
- Install Terera and activate your key while online.
- Run any network monitor — Windows' own Resource Monitor (
resmon, Network tab), Wireshark, or your corporate proxy logs. - Use the app hard: paste a sensitive script, speak, go off script, check your run summary.
- After setup, check that speech recognition and reading continue with the network disconnected. When online, model setup or repair and licence validation can generate requests.
If you observe traffic that seems unrelated to these purposes, contact security@terera.app. Include the app version and what you observed so we can investigate.
The website makes the same promise
terera.app sets no cookies, runs no analytics or tracking scripts, and shows no consent banner, because there's nothing to consent to. Every font and script is served from our own origin. The one exception is the checkout page, where Paddle — the merchant of record — sets the cookies its secure payment flow needs, and only there. Cloudflare keeps standard server logs; the privacy policy says exactly what that means.
For IT and procurement
| Runtime | Windows 10/11 desktop app (Electron), per-user install, no admin rights required. macOS next. |
| Speech processing | On-device (sherpa-onnx streaming recogniser). No cloud speech API, no audio retention. |
| Data at rest | Scripts, settings, stats in the user's %APPDATA% profile. Uninstall offers full data deletion. |
| Egress allowlist | Model downloads start at huggingface.co and can redirect to its file-delivery hosts. Licensing uses api.keygen.sh. Inspect setup requests when preparing a corporate allowlist. |
| Code signing | Authenticode via Azure Trusted Signing — publisher TERKON PTY LTD. The download page serves the current installer's SHA-256 in a response header. |
| Accounts & auth | None. A licence key, three machines, no passwords to breach. |
| Licensing at scale | Team and volume licensing: sales@terera.app. |
Statements on this page describe the current shipping version and are kept in lockstep with releases. If a future feature changes any of it — an optional update checker, for example — this page and the privacy policy change first, visibly.