HelloFace SDK 0.8.11 delivers offline face recognition across Android, iOS, HarmonyOS and Linux

Face recognition that never leaves the device

A project called HelloFace has released a software development kit that performs face recognition entirely on the hardware it runs on. According to a post on dev.to, HelloFace SDK version 0.8.11 offers native integration paths for Android, iOS, HarmonyOS, Windows and Linux, and processes all face images and recognition features locally rather than relaying them to a remote recognition service.

The feature set described by the project covers the full pipeline an access-control or identity-verification product typically needs: face detection and feature extraction, guided face capture, both action-based and passive liveness detection, 1:1 face verification, and 1:N and M:N face search backed by a local face database.

Linux and embedded support

For Linux, the SDK provides x86_64 and ARM64 binaries, a public C ABI, command-line examples and a customizable desktop demo. Applications can integrate the SDK without any desktop environment, though the bundled graphical demo requires X11 or XWayland.

The embedded Linux story goes further: the demo supports screen rotation, controls designed for touch input, and RGB/IR paired-search integration for supported hardware. The project cautions that camera configuration, calibration and performance should be validated on the actual target device rather than assumed from demo behaviour.

Mobile integration and the Google Play demo

On the mobile side, HarmonyOS developers integrate through ArkTS interfaces and a native HAR package, while Android and iOS each have their own integration guides and demos. Public demo code and platform guides live on GitHub under the FaceSDKPro/HelloFace repository, but the recognition engine itself ships only as prebuilt binaries, so the core cannot be inspected or rebuilt from source.

The Android demo app has now cleared Google Play's publication review, which checks apps against safety and privacy policies. dev.to is careful to separate the two claims here: the SDK processing faces on-device says nothing about the app's overall data practices, and readers are pointed to the Play Store's Data safety section for the developer's current disclosures.

Beta status comes with real restrictions

Version 0.8.11 is a Beta release intended for testing, and the stated limitations are worth noting before anyone builds on it. The current models do not support children under 12 and must not be used in financial scenarios. Face search depends on good-quality camera images, and outdoor deployments require cameras with wide dynamic range above 100 dB. Windows and Linux desktop and camera workflows still need validation with real cameras, and developers shipping mobile apps must handle their own application signing.

Why it matters

Most face recognition products route images through a cloud API, which creates two dependencies: connectivity and trust in the provider's data handling. An SDK that keeps the entire pipeline on-device removes both, which matters for embedded hardware in the field, kiosks, access control and any deployment where biometric data leaving the premises is a privacy or compliance problem rather than a feature. A cross-platform C ABI with ARM64 support also makes the toolkit viable on boards and gateways that could never host a cloud round-trip in the first place.

Two caveats temper the appeal. The binary-only core means adopters must trust the vendor's accuracy and privacy claims without being able to audit the engine. And the explicit restrictions — no children under 12, no financial use, and camera workflows not yet validated on desktop — mark this as early-stage tooling rather than a production component. For developers evaluating offline biometrics, the public demo code and per-platform guides make it a reasonable starting point for testing those claims on real hardware.