Integrate Claude API into Your Mobile App

Typical scenario: you're building a mobile chat assistant in Swift or Kotlin and choosing between OpenAI and Anthropic. Anthropic's Claude API offers up to 200k token context (claude-3-5-sonnet), native vision support, and excellent Russian language quality. We integrate Claude into your app turnkey: from backend-proxy architecture to final streaming testing. Our expertise — over 30 AI integration projects — helps reduce time-to-market by 2-3x compared to in-house development.

Secure Key Management

The Anthropic API key (sk-ant-...) must never be stored on the client. Rule: key stays on backend only. The mobile client communicates with your proxy server, which adds the x-api-key header and forwards requests to api.anthropic.com. Proxy architecture: any backend — Laravel, FastAPI, Cloudflare Worker. A minimal Cloudflare Worker implementation takes ~30 lines and handles both regular and streaming requests. Workers cold start — 5–10 ms, latency unnoticeable.

On the mobile client: JWT authentication to the proxy. The proxy validates the token, applies rate limiting (e.g., 20 requests/minute per user), and logs input_tokens/output_tokens for cost analytics. We ensure the key never leaves the backend.

Why Messages API Differs from OpenAI

The Anthropic Messages API differs from OpenAI Chat Completions in several ways:

• System prompt: separate system field, not an element in messages. Best practice: keep system context in system, not in messages[0] with role: "system".
• Roles: only user and assistant (no system in messages).
• No function_calling — use tools with input_schema in JSON Schema format.

{
  "model": "claude-haiku-4-5",
  "max_tokens": 1024,
  "system": "You are a mobile app assistant...",
  "messages": [
    {"role": "user", "content": "Explain this document"},
    {"role": "assistant", "content": "Sure, ..."},
    {"role": "user", "content": "What does clause 3 mean?"}
  ]
}

Streaming on Mobile: How to Speed Up Responses

Claude API supports SSE streaming with stream: true. The format differs slightly from OpenAI: content_block_delta event carries delta.text — that's one token; message_stop signals end of stream. On iOS, parse via URLSessionDataDelegate; on Android, use OkHttp EventSource. Delta events arrive every 10–50 ms during active generation. Buffer before UI updates: update @Published var streamText not at each event, but via a Throttle publisher (iOS) or distinctUntilChanged + debounce (Android Flow).

Comparison of streaming Claude vs OpenAI:

How to Analyze Images via Claude on Mobile?

Claude 3+ natively supports images in messages. Format:

{
  "role": "user",
  "content": [
    {
      "type": "image",
      "source": {
        "type": "base64",
        "media_type": "image/jpeg",
        "data": "<base64>"
      }
    },
    {"type": "text", "text": "What is in this photo?"}
  ]
}

On mobile: compress image before sending. JPEG quality 70, max size 1568×1568 (API limit). Resize + compress via UIGraphicsImageRenderer (iOS) or Bitmap.createScaledBitmap + compress (Android). Token savings of 5–10x vs sending RAW.

Conversation Management and RAG

Claude handles 200k tokens, but for a mobile chat this is overkill and expensive. In practice, a sliding window of the last 20 messages is enough. For specialized apps (legal assistant, medical reference) — RAG (Retrieval Augmented Generation): store documents in a vector DB on the backend, augment the system prompt with relevant fragments per request. This doesn't grow history size but provides access to a large knowledge base. Learn more about RAG in Anthropic's documentation.

Handling Anthropic API Errors

529 Overloaded — servers overloaded, apply exponential backoff. 400 with error.type = "invalid_request_error" — usually max_tokens exceeded or invalid content format. 401 — wrong key on proxy. Log all errors with request ID (x-request-id) — needed for Anthropic support.

Case: legal assistant for a B2B app. Used claude-3-5-sonnet, contract analysis. User photographs a contract page, the assistant highlights key terms and risks. Image resized to 1200px on long side, JPEG 80. Average request: 2400 input tokens (image ~1800 + text 600) + 800 output. Streaming — first words appear in 350 ms. Users don't notice latency with streaming vs a "blank screen for 4 seconds" without it.

Claude Model Comparison

Model selection depends on the scenario: for simple chat use Haiku, for complex analysis — Sonnet or Opus. We help choose the optimal model and set up fallback to reduce costs.

What's Included in the Work

• Backend-proxy architecture (Cloudflare Worker / Laravel / FastAPI)
• Messages API integration with streaming and vision
• Conversation management (sliding window, RAG if needed)
• Token logging and error handling
• Deployment and support documentation
• Testing with TestFlight / Firebase App Distribution

Timelines and Cost

Basic integration with streaming, conversation context, and backend proxy — 3–5 business days. With image support and RAG — 1–2 weeks. Cost is calculated individually. Contact us for a project assessment — we'll prepare a proposal within a day. You can also order an architecture consultation — its cost will be deducted from the main project.

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.