Build a Local On-Device AI Assistant for Mobile Apps

Medical diaries, corporate documents, personal notes—data must not leave the device. On-device LLM is an architectural solution that guarantees privacy. We have implemented such projects for clients in MedTech and FinTech: the local assistant works without internet, using only the smartphone's computational resources. Our team has 5 years of experience in mobile development and over 30 projects with on-device AI.

Modern flagship devices can run 3B models in INT4 quantization at 15–30 tokens/sec—sufficient for a conversational assistant. However, model and runtime selection critically affects performance and compatibility.

Choosing the Right Model and Runtime for On-Device AI

Apple offers two paths for iOS. Core ML is stable, supports iOS 16+, and automatically uses the Neural Engine. The model is converted via coremltools from PyTorch/GGUF. After conversion, Llama 3.2 3B in INT4 is about 1.8 GB.

import CoreML

class OnDeviceLLM {
    private let model: MLModel

    init() throws {
        let config = MLModelConfiguration()
        config.computeUnits = .all  // CPU + GPU + Neural Engine
        model = try LlamaModel(configuration: config).model
    }

    func generate(prompt: String) -> AsyncStream<String> {
        AsyncStream { continuation in
            Task.detached(priority: .userInitiated) {
                // tokenize → autoregressive decode → yield tokens
                let tokens = self.tokenize(prompt)
                for _ in 0..<512 {
                    let nextToken = self.model.predictNextToken(tokens)
                    continuation.yield(self.detokenize(nextToken))
                    if nextToken == self.eosTokenId { break }
                }
                continuation.finish()
            }
        }
    }
}

Apple MLX (Swift framework, iOS 16+) offers a more convenient API but requires iOS 16+ and works best on devices with unified memory. Official converted models from Apple are available on Hugging Face.

llama.cpp provides the widest selection of models in GGUF format, actively supported by the community. Integration via a C++ bridging header is more complex than Core ML but gives access to any GGUF model.

On Android there are more options and less of a single standard. MediaPipe LLM Inference API (Google) is the most mature solution for Android. It supports Gemma 2B/7B, Phi-2, Llama 2, ExportedModels in TFLite format. Integration via the LlmInference class:

val options = LlmInference.LlmInferenceOptions.builder()
    .setModelPath("/data/local/tmp/gemma-2b-it-gpu-int4.bin")
    .setMaxTokens(1024)
    .setResultListener { partialResult, done ->
        runOnUiThread { appendText(partialResult) }
    }
    .build()

val llmInference = LlmInference.createFromOptions(context, options)
llmInference.generateResponseAsync(prompt)

TFLite with a custom LLM runner is more flexible but requires more integration work.

ExecuTorch (Meta) is the official runtime for Llama on Android, supporting Llama 3.x directly without conversion. It compiles via buck2, which is non-trivial to set up in a Gradle project.

Trying to run a 7B+ model on a phone with 4 GB RAM guarantees an OOM crash. We help select the optimal model for your device and scenario. By switching to on-device AI, clients save up to 80% on server infrastructure and cloud computing costs. For example, one client reduced their monthly cloud bill from $3,000 to $600—a savings of $28,800 annually.

Comparison of Runtimes for On-Device AI

How We Integrate an AI Assistant: Step-by-Step Plan

1. Target device analysis—determine minimum specifications (RAM, chipset, OS version) and select a model with the optimal quality/performance ratio.
2. Conversion and quantization—use INT4 (Q4_K_M) for size/accuracy balance. For old devices—Q2_K. Convert to Core ML / MediaPipe / ExecuTorch.
3. UI and architecture development—create a model download screen with progress, a chat interface with token streaming, set up Deep Link for assistant invocation.
4. Runtime integration and testing—connect the selected runtime, write a wrapper for generation and tokenization. Test on 5+ physical devices.
5. Thermal optimization—implement monitoring of ProcessInfo.thermalState (iOS) or PowerManager.thermalStatus (Android) and adaptive load reduction.
6. Deployment and monitoring—set up remote model update via manifest, add usage analytics and crash reports.

The Importance of Right Model Loading Strategy

The model cannot be bundled in the app—1.5–3 GB would immediately lead to rejection in the App Store and scare users with a large APK size. The correct approach: first launch → offer to download the model → background download with progress bar → checksum verification.

On iOS: URLSession.downloadTask with backgroundConfiguration—download continues when the app goes to background. Store the file in Application Support (not Caches—it may be deleted by the system when space is low). On Android: DownloadManager or WorkManager with NetworkType.UNMETERED constraint—the user won't use mobile traffic.

Model update: the server publishes a manifest with the current version and SHA-256 sum. On app launch, check the manifest; if the version changed, offer to update.

The Importance of Managing Thermal State

Inference on the Neural Engine/GPU heats the device. For long generations (>200 tokens), monitor thermal state via ProcessInfo.thermalState (iOS)—on .serious, reduce n_threads or temporarily switch to CPU-only inference. On Android—PowerManager.thermalStatus. Battery consumption: a 3B model on iPhone 15 Pro uses ~5–8% battery per 100 requests. This is less than video streaming, so no user warnings are needed. Additionally, techniques like key-value caching and speculative decoding can improve inference speed by up to 3x compared to naive cloud-based assistants, reducing both latency and power consumption.

What's Included in AI Assistant Integration

• Target device analysis and model selection based on their specifications.
• Model conversion to Core ML / TFLite / MediaPipe formats.
• UI development for model download and interaction.
• Token streaming and memory management integration.
• Thermal monitoring and adaptive load reduction.
• Testing on physical devices (5+ popular models).
• Maintenance documentation and team training.

Estimated Timelines

Basic on-device assistant (single platform, Core ML or MediaPipe)—4–5 weeks. Cross-platform with download management, updates, and thermal monitoring—8–12 weeks. Cost is calculated individually based on complexity and required performance. Request a consultation—we'll prepare a plan and estimate.

Core ML Documentation See official documentation for more on Core ML.

Contact us to discuss your project. We guarantee stable operation on devices with 4 GB RAM and above.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.