Mobile AI Assistant with Claude (Anthropic): 200K Context, Streaming, Vision

Why Claude for a Mobile AI Assistant?

Imagine a user uploading a 100-page contract PDF and wanting to ask questions about it. Standard assistants with 8K context fail — you have to split the document, losing integrity. Claude solves this: its 200K token context window allows loading the entire document and answering without an RAG pipeline. We have over 5 years of experience developing mobile solutions and guarantee quality AI integration into your app. We evaluate your project in 1–2 days — just reach out to us.

Anthropic Messages API: Structure and Peculiarities

The Anthropic API is structurally similar to OpenAI, but with important differences. The system prompt in Claude is a separate system parameter, not a message with role system in the messages array. This is critical: trying to pass the system prompt inside messages degrades instruction-following quality.

struct AnthropicRequest: Encodable {
    let model: String          // "claude-3-5-sonnet-20241022"
    let maxTokens: Int         // mandatory, no default
    let system: String         // system prompt — separate
    let messages: [Message]
    let stream: Bool

    enum CodingKeys: String, CodingKey {
        case model, system, messages, stream
        case maxTokens = "max_tokens"
    }
}

max_tokens in the Anthropic API is a mandatory parameter with no default. If you forget to pass it, the API returns a 400 error. This differs from OpenAI, where max_tokens is optional.

Authentication: the x-api-key header (not Authorization: Bearer). API versioning via anthropic-version: 2023-06-01. Without this header — 400 Bad Request.

How to Implement Streaming from Claude on iOS?

Claude supports streaming via Server-Sent Events. The stream structure differs from OpenAI: events content_block_start, content_block_delta, content_block_stop, message_delta — each carries its own fields.

Here is a step-by-step implementation on iOS:

1. Initialize URLSession and create a request with headers.
2. Use bytes (AsyncSequence) to read the stream.
3. For each line, check the prefix "data: ".
4. Decode JSON into a struct with a type field.
5. For content_block_delta, extract text and update UI on the main thread.

for try await line in response.bytes.lines {
    guard line.hasPrefix("data: ") else { continue }
    let jsonString = String(line.dropFirst(6))
    guard jsonString != "[DONE]" else { break }

    if let data = jsonString.data(using: .utf8),
       let event = try? JSONDecoder().decode(StreamEvent.self, from: data),
       event.type == "content_block_delta" {
        let delta = event.delta?.text ?? ""
        await MainActor.run { self.appendText(delta) }
    }
}

It is important to handle all event types, not just content_block_delta — message_delta contains stop_reason (e.g., max_tokens), which you should show to the user.

Advantages of a Large Context on Mobile

200K tokens — roughly 150,000 words or ~500 pages of text. For a mobile assistant, this means working with full documents without an RAG pipeline. The user attaches a contract PDF — you can pass it entirely in the context and ask questions.

The downside: large context = long time-to-first-token. With 50K tokens in the request, the first response token can take 3–5 seconds even on a good connection. On mobile, you need a progress indicator that appears immediately, before the first token, otherwise the user thinks the app is frozen.

Cost also grows linearly with context — for apps with user billing, it is important to consider when designing a token counter UI. Claude 3.5 Sonnet processes 200K token context 2x faster than GPT-4o with the same volume, making it ideal for mobile scenarios with long conversations. Token savings can reach 40% compared to competitors, and overall infrastructure costs are lower thanks to native long-context support.

What Does a 200K Token Context Give to a Mobile User?

Let's compare key parameters of Claude 3.5 Sonnet and GPT-4o on mobile:

Claude wins on context volume and speed with large datasets. The API is stricter, but this reduces errors when configured correctly.

Vision: Sending Images to Claude

Claude 3.5 Sonnet supports images via base64 in a content block:

let imageContent = ContentBlock(
    type: "image",
    source: ImageSource(
        type: "base64",
        mediaType: "image/jpeg",
        data: imageBase64
    )
)

Limitation: maximum 20 images per request, each up to 5 MB. On mobile, compress the image before sending to a reasonable size — UIGraphicsImageRenderer or BitmapFactory.Options with inSampleSize.

Process and What's Included

Key parameters to clarify: whether document support (PDF, images) is needed, expected conversation volume, whether a server-side proxy is needed (yes — mandatory, API key is not stored in the app).

What's included in the work:

• Documentation for Claude API integration on your platform
• Configured proxy server with authentication
• Ready Swift/Kotlin client for streaming
• Instructions for App Store Review (handling ATT, In-App Purchase requirements)
• 2 weeks of technical support after launch

Implementation: Anthropic API client → streaming UI → history management with 200K limit → optional file handling.

Estimated Timelines

Basic text assistant — 1–2 weeks. With document, image support, and server-side proxy — 3–4 weeks. Exact timelines depend on your stack and specifics. Request a free consultation — we'll show a demo version and calculate the cost. Over 50 AI integration projects under our belt. Contact us for a free project evaluation.

Anthropic API documentation: https://docs.anthropic.com/en/api

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.