flutter_concrete 0.1.0

Concrete ML FHE client for Flutter. Native TFHE-rs encryption/decryption via Dart FFI with quantization support.

flutter_concrete

A Flutter FFI plugin that brings Concrete ML FHE (Fully Homomorphic Encryption) to mobile apps. The native cryptographic operations — key generation, encryption, and decryption — run entirely on-device via TFHE-rs, with no server-side private key material.

The Rust library builds automatically during flutter build thanks to Cargokit — no manual build scripts or precompiled binaries required.

How it works

Your App                              flutter_concrete
───────                               ────────────────
                                           │
Load client.zip from assets ──────►  setup(zipBytes, storage)
                                       parse serialized_processing.json
                                       restore or generate keys
                                     ◄── isReady = true
                                           │
Get serverKey ◄────────────────────  serverKey / serverKeyBase64
Upload to your server (your code)          │
                                           │
Float32 features ──────────────────►  quantizeAndEncrypt()
Uint8List ciphertext ◄──────────────       │
Send to server (your code)                 │
Receive result (your code)                 │
Uint8List encrypted result ────────►  decryptAndDequantize()
Float64 class scores ◄──────────────       │
Interpret scores (your code)

The server performs ML inference on encrypted data — it never sees plaintext inputs or predictions.

Installation

Add to your pubspec.yaml:

dependencies:
  flutter_concrete:
    git:
      url: https://github.com/afonsomota/flutter_concrete.git

Or as a local path dependency:

dependencies:
  flutter_concrete:
    path: ../flutter_concrete

Prerequisites

• Rust toolchain — install via rustup
• iOS targets: rustup target add aarch64-apple-ios aarch64-apple-ios-sim x86_64-apple-ios
• Android targets: rustup target add aarch64-linux-android armv7-linux-androideabi x86_64-linux-android

Usage

import 'package:flutter_concrete/flutter_concrete.dart';

// 1. Implement KeyStorage (e.g. wrapping flutter_secure_storage)
class MyKeyStorage implements KeyStorage {
  @override
  Future<Uint8List?> read(String key) async { /* ... */ }
  @override
  Future<void> write(String key, Uint8List value) async { /* ... */ }
  @override
  Future<void> delete(String key) async { /* ... */ }
}

// 2. Create client and set up from Concrete ML's client.zip
final client = ConcreteClient();
final zipBytes = await loadClientZipFromAssets(); // your asset loading
await client.setup(
  clientZipBytes: zipBytes,
  storage: MyKeyStorage(),
);
// First call generates keys (~10-60s on mobile), subsequent calls restore.

// 3. Get server key to upload to your backend
final serverKey = client.serverKey;       // Uint8List
final serverKeyB64 = client.serverKeyBase64; // String (cached)
// Upload to your server however you want

// 4. Encrypt features
final ciphertext = client.quantizeAndEncrypt(featureVector);
// Send ciphertext to your server for FHE inference

// 5. Decrypt server response
final scores = client.decryptAndDequantize(encryptedResult);
// scores is Float64List — apply argmax for classification

API

ConcreteClient

Method	Description
Future<void> setup({clientZipBytes, storage})	Parse client.zip, generate/restore keys
void reset()	Clear state so setup() can be called with a different model
bool get isReady	True after setup() completes
Uint8List get serverKey	Raw evaluation key bytes (throws before setup)
String get serverKeyBase64	Base64-encoded server key (cached)
Uint8List quantizeAndEncrypt(Float32List)	Quantize + FHE encrypt
Float64List decryptAndDequantize(Uint8List)	FHE decrypt + dequantize

KeyStorage (abstract — you implement this)

Method	Description
Future<Uint8List?> read(String key)	Read stored bytes, or null
Future<void> write(String key, Uint8List value)	Persist bytes
Future<void> delete(String key)	Delete entry

Compatibility

• Concrete ML: Accepts standard client.zip from FHEModelDev.save()
• TFHE-rs: Git revision 1ec21a5 (matching concrete-ml-extensions 0.2.0)
• Parameter set: V0_10_PARAM_MESSAGE_2_CARRY_2_KS_PBS_GAUSSIAN_2M64
• Serialization: bincode (ciphertexts), Cap'n Proto (evaluation keys)
• Platforms: iOS, Android

Known limitations

1. Hardcoded eval key topology — key generation produces 4 BSKs and 8 KSKs matching a specific Concrete ML circuit. Future: parse client.specs.json for dynamic key topology.

2. uint8 input / int8 output only — matches 8-bit quantization. Other bit widths are not yet supported.

3. Single input/output tensor — assumes one input and one output tensor per circuit.

4. No precompiled binaries — requires Rust toolchain on the build machine.

License

MIT