How do I install the app?

There are two paths, and your organization will have chosen one.

Through your MDM (managed deployment)

Your administrator assigns SecureEdge to your device from Apple Business Manager or the App Store through the MDM (Intune, Jamf, Ivanti, Workspace ONE, and others). The app appears on your device with its managed configuration already applied, and Settings inside the app shows your organization's name. Model bundles usually arrive the same way. If the app installed but Settings says No managed configuration, the profile has not reached the device yet; ask your administrator to check the assignment.

Through the App Store or a pilot build (self-managed)

Install the app as you would any other. Without a managed configuration it runs with development defaults and says so in Settings. You then add a model: in the App Store build, download one from the catalog (see Downloading a model), or in any build, import a bundle yourself through the Files app (see Importing a model bundle). A pilot build we send you may already include a model.

Which devices are supported?

The floor is set by the language model's memory needs, not by iOS. iPhone 15 or later on iOS 26 is the minimum. iPhone 15 Pro or later (8 GB) is recommended and is required for the larger Gemma 4 E4B and the vision-capable variant. Any iPad on iPadOS 26 with 8 GB works the same way. The app checks each model's minimum-RAM requirement against the device and explains, in the model picker, why a model cannot be loaded. See the full requirements table.

Why is dictation unavailable for my language?

SecureEdge uses Apple's speech recognizer with on-device recognition required. It will never fall back to Apple's server-based recognizer, because that would send audio off the device. Apple supports on-device recognition for a subset of languages, and the language assets must be present on the device. If dictation is greyed out or reports that on-device recognition is not available:

Open iOS Settings ▸ General ▸ Keyboard ▸ Dictation and make sure dictation is enabled and the language you want is listed. iOS downloads the on-device language pack from there. This is an iOS system download, not something the app does.
Confirm the language is one Apple supports for on-device recognition. If it is not, dictation for that language is not possible in SecureEdge by design.
Check that your administrator has left allowDictation enabled. If policy disables it, the app does not request microphone access at all.
Make sure the app has microphone and speech-recognition permission in iOS Settings ▸ SecureEdge.

The app's first release recognizes English (United States). Additional languages depend on Apple's on-device support and will be added as they are validated.

How do I download a model?

Model downloads are available in the App Store build only. The managed build your organization distributes through MDM contains no networking code, so models reach it as bundles instead (see Importing a model bundle). Nothing is downloaded automatically: the app contacts the catalog only when you open the download list, and fetches a model only when you tap Download.

In Chat, tap the model card at the top of the screen, then Models.
Under Available to download, find the model you want. The list is fetched from the catalog when you open it, so the device needs a connection at that moment; nothing else in the app depends on one.
Tap Download. The app shows the file size before it starts. Models are several gigabytes, so use Wi-Fi unless your cellular plan can absorb it.
You can leave the app; the download continues in the background, and a local notification tells you when it finishes if the app is not in front.
When the download completes, the app streams the file through SHA-256 and verifies the manifest signature. A file that fails either check is deleted, and you can try again. A file that passes appears under your installed models.

To remove a downloaded model, open the same Models screen and tap Remove on the installed model's card. That deletes the file from the device; you can download it again later.

If Available to download is missing or downloads are refused, your administrator has set allowModelDownloads to false. Only the administrator can change that; ask for a bundle instead. An administrator may also point the app at the organization's own catalog with modelCatalogURL, in which case the models listed are the ones your organization has chosen to offer.

The models offered from our catalog are Google's Gemma 4 E2B and Gemma 4 E4B in LiteRT-LM format, redistributed under the Gemma Terms of Use. Downloading one is subject to those terms. Each download is a plain HTTPS request that carries no account, device identifier, or content of yours, and we keep no logs of it; the privacy policy has the details.

How do I import a model bundle?

A model bundle is a folder containing manifest.json and the .litertlm model file it names. Apart from a download you start in the App Store build, the app never fetches models; it looks in two places:

Models/ inside the app itself, for pilot builds that ship with a model.
The app's Documents folder on the device, visible in the Files app under On My iPhone ▸ SecureEdge ▸ Models. MDM asset pushes, Finder copies over USB, and the in-app import all land here.

To import one yourself:

Get the bundle folder onto the device: AirDrop it, copy it with Finder while the phone is connected by USB, or save it from any Files provider your organization allows.
In Files, move the folder into On My iPhone ▸ SecureEdge ▸ Models. Alternatively, open the model picker in the app and choose Import bundle from Files, then pick the folder.
In the model picker, tap Rescan if the model is not already listed. The app verifies the manifest signature and the file hash and shows the model with its status.

Bundle folders are large (several gigabytes). Copy them over USB or AirDrop rather than through a cloud storage provider, and remember that allowFileImport must be enabled by policy for the in-app import to appear.

Why is a model blocked by policy?

The picker shows the reason beside each model that cannot be loaded. The common ones:

Not in allowed list. Your administrator's allowedModels names specific model ids and this one is not among them. Only the administrator can change that.
Signature not trusted. The manifest was signed with a key that is not in trustedSigningKeys. Either the bundle came from somewhere other than your organization, or the key has been rotated. Ask for a bundle signed with the current key.
Unsigned. The manifest has no signature and requireSignedModels is true (the default). Sign it with seadmin sign.
Hash mismatch. The model file does not match the SHA-256 in the manifest. The copy is corrupt or was altered. Delete it and copy it again.
Insufficient memory. The device has less RAM than the manifest's minimum. Choose a smaller model or a newer device.

Every refusal is written to the audit log with its reason, so your security officer can see it too.

The app says it is in restricted mode

A managed configuration reached the device but could not be parsed: a wrong type, an unknown retention value, a malformed key. The app fails closed and disables dictation, image input, imports, export, and model loading until a valid configuration arrives. Nothing is lost. Ask your administrator to regenerate the configuration with seadmin mdm-config and push it again.

How do I verify the audit log?

On the device, open Settings ▸ Audit log and tap Verify chain. The app walks the entire hash chain from the genesis entry and reports either that it is intact, with the entry count and head hash, or the sequence number of the first entry where the chain breaks. The verification itself is added to the log.

To verify independently, export the log as JSON Lines (requires allowAuditExport), move the file to a workstation, and run seadmin audit-verify against it. The tool recomputes every hash without needing the device. A verified export plus the head hash noted on the device is the evidence a security review usually wants.

Where did my data go?

Your administrator sets a retention period: the current session, 24 hours, 30 days, or indefinite. When it elapses, chats, transcripts, and imported documents older than the limit are purged automatically and a retentionPurge entry is added to the audit log. This is by design and cannot be changed from inside the app. If you need something to outlive the retention window, export it (where allowExport permits) before it expires.

Does the app ever go online?

Only when you ask it to, and only in the App Store build. The managed build your organization distributes contains no networking code; a test in our build fails if any is added, and the compiled binary links no networking symbols. It cannot download models, check for updates, send diagnostics, or reach any server. The App Store build has exactly one network feature, model downloads, and it connects only when you open the download list or tap Download (see Downloading a model). Neither build sends prompts, documents, diagnostics, or anything else off the device, and neither checks for updates; app updates come through the App Store or your MDM like any other app. With a model installed, airplane mode makes no difference to how either build works. The details are on the security overview.

Contact

Email babs@wanyekitech.com. When writing about a problem, please include:

Your device model and iOS version (Settings ▸ General ▸ About).
The app version and whether Settings shows a managed configuration.
The model id in use, if the problem involves a model.
What you expected and what happened instead. Screenshots help; remember they are recorded in the audit log.

Please do not include controlled or sensitive information in a support email.

Evaluating SecureEdge for a unit or a team? Request a pilot and tell us which MDM you run.