Implementing an AI Chatbot in a Mobile App: From Idea to Production

Implementing an AI Chatbot in a Mobile App

Integrating GPT-4o or Claude into a mobile chat isn't just "connect an SDK and you're done." The real complexity begins after the first working request: managing dialogue context, displaying streaming generation without UI jank, handling network issues on weak signals, and storing chat history between sessions without leaking personal data. We are a team with 8 years of mobile development experience, having delivered 40+ projects with AI features. We offer turnkey AI chatbot integration: from model selection to store publication. We'll evaluate your project in 1–2 days.

How to Manage Dialogue Context Without Losing Quality?

All LLMs are stateless. Each request to OpenAI, Anthropic, GigaChat, or YandexGPT sends the full dialogue history. This means: storing and truncating context is your job. A naive implementation after 20 messages can increase token cost by 3–4 times, and with a 128k context, you might wait 30+ seconds for a response.

A practical solution is a sliding window with summarization:

class ConversationManager {
    private var messages: [ChatMessage] = []
    private let maxMessages = 20
    private let summaryThreshold = 15

    func addMessage(_ message: ChatMessage) {
        messages.append(message)
        if messages.count > summaryThreshold {
            Task { await compressSummary() }
        }
    }

    private func compressSummary() async {
        // Take messages before threshold, summarize with a separate LLM request
        let toCompress = Array(messages.prefix(10))
        let summary = try? await llmClient.summarize(messages: toCompress)
        if let summary {
            messages = [ChatMessage(role: .system, content: "Context: \(summary)")] +
                       Array(messages.suffix(10))
        }
    }
}

The system prompt is a separate story. It must always remain the first message. Do not touch it when compressing context.

Streaming Generation and UI

Users shouldn't wait for a full response. Streaming via SSE is the standard for all modern LLM APIs. On iOS:

// Update SwiftUI View via @Published
class ChatViewModel: ObservableObject {
    @Published var streamingText = ""

    func streamResponse(for prompt: String) {
        streamingText = ""
        Task {
            for try await chunk in llmClient.stream(prompt: prompt) {
                await MainActor.run {
                    streamingText += chunk
                }
            }
        }
    }
}

On Android with Compose, use StateFlow<String> collected with collectAsState(). A typical mistake: calling notifyDataSetChanged() or recreating a RecyclerView adapter on each chunk — this causes visible flickering. Update only the last message's text, not the entire list.

Offline Mode: On-device vs Cloud

For basic scenarios, use Apple Intelligence API (iOS 18+) or SmartReply from ML Kit. For more complex ones, use llama.cpp via Metal/CoreML. We guarantee correct operation with any stack.

Storing Dialogue History

Chat history contains personal data. Use SQLite/Core Data with encryption via SQLCipher or iOS Data Protection. Do not store history in UserDefaults — it syncs to iCloud without encryption. On Android, use Room with EncryptedSharedPreferences for encryption keys.

Cleanup strategy: auto-delete dialogues older than N days, or explicit deletion on user request — this is a requirement of GDPR and Russian Federal Law 152-FZ.

What's Included in the Work

• Architecture: LLM provider selection, on-device vs cloud, authorization scheme.
• Backend proxy with rate limiting, caching, logging.
• ConversationManager: sliding window, summarization, system prompt.
• Chat UI: bubble layout, streaming, typing indicator, reaction buttons.
• History storage: encryption, auto-cleanup, export.
• Moderation API: input and output filtering.
• Testing edge cases: network loss, long responses, concurrent requests.
• Documentation: README, flow diagram, deployment instructions.
• Support: 2 weeks after handover (consultation, bug fixes).

Typical Production Issues

Repetitive responses. GPT sometimes gets stuck on a pattern. Parameters presence_penalty: 0.6 and frequency_penalty: 0.3 reduce the likelihood. If stuck, implement client-side detection: if the last 3 bot messages contain >60% identical n-grams, reset the context.

Timeout on poor network. LLMs can generate slowly. Default URLSession timeout is 60 seconds, which is too short for long streaming responses. Set timeoutIntervalForResource: 120 and add an extra progress indicator "thinking..." after 5 seconds of no first chunk.

Moderation. OpenAI Moderation API before sending user input is mandatory for public apps. One POST /v1/moderations is cheaper than dealing with an App Store Review complaint.

Process

1. Architecture design: LLM provider selection, on-device vs cloud, authorization scheme.
2. Backend proxy development with rate limiting.
3. ConversationManager implementation with context management.
4. Chat UI: streaming, bubble layout, typing indicator.
5. Dialogue history with encryption.
6. Edge-case testing: network loss during generation, very long responses, concurrent requests.

Timeline Estimates

A simple chatbot with one LLM provider and no history: 5–7 days. A full-featured chatbot with history, context compression, offline mode, and moderation: 3–5 weeks. Cost is calculated individually. Contact us for a consultation and project estimate within 1–2 days.

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.