YandexGPT Integration for Mobile Apps: AI Assistant Turnkey

Integrating an AI assistant based on YandexGPT into a mobile app for the Russian market costs from $2,000 and achieves up to 30% savings compared to in-house development. Our team has over five years of experience and has integrated YandexGPT into more than 15 projects, from fintech to e-commerce. We guarantee compliance with App Store Review Guidelines, secure key handling via a server proxy, and full support at all stages. For a typical e-commerce app with 10,000 monthly active users, the monthly API cost is approximately $80 for Lite. Get a consultation to assess your project.

Integrating an AI Assistant with Voice Control

For voice input/output in a Russian app, Yandex SpeechKit offers the best quality of Russian speech among available services. 90% of our clients choose Lite over Pro for their mobile assistants. SDKs for iOS and Android are available via CocoaPods and Maven respectively. STT via WebSocket: wss://stt.api.cloud.yandex.net/speech/v3/stt:streamingRecognize — streaming recognition with partial results. TTS via REST with voice selection (alena, filipp, jane — SSML is supported). For a voice assistant, we recommend combining SpeechKit with YandexGPT: recognition → response generation → synthesis.

A Server Proxy Is Critical for YandexGPT

The IAM token lives for 12 hours and must not be stored in the app. Direct use of the IAM token violates Section 5.1 of the App Store Review Guidelines (Privacy). A server proxy also allows caching responses and managing limits. We use a server proxy: the app sends requests to your backend, which signs them with a service account. This prevents key leakage and gives you control over load.

Yandex Foundation Models API

YandexGPT is available via the Yandex Cloud Foundation Models API. Base URL: https://llm.api.cloud.yandex.net/foundationModels/v1/completion. Authentication uses an IAM token (for user apps) or a service account API key (for server proxy). The IAM token lives for 12 hours and requires refreshing — it is not used directly on the mobile client.

Request structure:

struct YandexGPTRequest: Encodable {
    let modelUri: String  // "gpt://{folder_id}/yandexgpt/latest"
    let completionOptions: CompletionOptions
    let messages: [YandexMessage]
}

struct CompletionOptions: Encodable {
    let stream: Bool
    let temperature: Double  // 0..1
    let maxTokens: String    // string, not number — API quirk
}

Important quirk: maxTokens is passed as a string, not a number. This violates the principle of least astonishment and periodically breaks automatically generated clients. modelUri is built as gpt://{folder_id}/{model_name}/{version}. folder_id is the Yandex Cloud folder identifier, it must be stored on the server, not in the app.

How YandexGPT Streaming Works in a Mobile App

To display responses in real time, streaming mode is needed (stream: true in the sync request). In this mode, the server returns chunked responses with partial results. Each chunk is a full JSON with accumulated text (not a delta, but the full text at each step). This is important: when rendering, you must replace the previous text with the new one, not append the delta as in OpenAI.

// Each chunk contains FULL text, not a delta
// Correct rendering:
func handleChunk(_ response: YandexCompletionResponse) {
    let fullText = response.result.alternatives.first?.message.text ?? ""
    DispatchQueue.main.async {
        self.currentMessage = fullText  // replace, not append
    }
}

YandexGPT supports synchronous mode (/completion) with a 60-second timeout and asynchronous mode (/completionAsync), where you first get an operation_id then poll for the result. Streaming is only available in synchronous mode.

Choosing Between YandexGPT Lite and Pro

For most mobile assistant tasks (helper, FAQ, text processing), Lite is sufficient — it handles about 80% of queries. Pro is justified for complex analytical tasks and working with long documents. The Embeddings API (/textEmbedding) is useful for semantic search in a local knowledge base — the text-search-query/latest model for queries, text-search-doc/latest for documents.

Process of Work

1. Setup Yandex Cloud: Create an account, create a service account, assign the ai.languageModels.user role, and set up a server proxy for secure credential storage. (Setup takes approximately 2-3 hours.)
2. Develop API Client: Build API client with proper streaming handling, implement UI for chat, add history management.
3. Integrate SpeechKit (optional): Add voice input/output using WebSocket STT and REST TTS.
4. Test and Deploy: Test on staging with 100+ test queries, then assist with App Store and Google Play publishing.

Typical Mistakes When Integrating

What's Included in the Work

• Requirements analysis and model selection (Lite/Pro)
• Yandex Cloud setup and server proxy deployment
• Streaming chat implementation with UI for iOS/Android
• SpeechKit integration (voice input/output)
• API documentation and authorization scheme
• Assistance with publishing on App Store / Google Play
• Post-launch support (2 weeks)

Timeline Estimates

Text assistant with streaming: 1–2 weeks. With voice via SpeechKit and server proxy: 3–4 weeks. Typically, a simple text assistant costs between $2,000 and $5,000, while a full voice assistant ranges from $5,000 to $10,000. Contact us: we will assess your project, propose an architecture and timeline. We work turnkey. Save time and money on setup and debugging — up to 30% compared to in-house development. Optimize your budget: delegate integration to professionals.

Links: YandexGPT API documentation and App Store Review Guidelines.

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.