Microsoft Bot Framework Integration for Mobile Apps

Integrating Microsoft Bot Framework into a Mobile App: NLP in Practice

Suppose you have a mobile app that needs to communicate with users in natural language — understand commands and respond across channels. If you build NLP from scratch, it will take six months and a budget comparable to a team of three developers. Microsoft Bot Framework v4 and Azure Bot Service provide a ready-made infrastructure: a dialog engine, integration with dozens of channels (Teams, Telegram, web), and a built-in NLP component CLU. However, integrating into a mobile app requires understanding architectural nuances. Let's break down the key points: Direct Line, tokens, dialog state, and avoiding Adaptive Cards. This saves up to 40% of the budget compared to custom development.

How to Securely Connect a Mobile App to Azure Bot Service?

Direct Line is the only channel for custom mobile clients (see Direct Line Protocol). The mistake we've seen in dozens of projects: developers hardcode DirectLineSecret into the app code. After reverse engineering the APK or IPA, the secret becomes public. This allows attackers to send requests on behalf of the bot. The solution is tokenization: the server generates a temporary token via /v3/directline/tokens/generate and passes it to the client. The token lives 30 minutes and can be refreshed via /v3/directline/tokens/refresh. Example in Swift:

// iOS: obtaining a token from the server and initializing a Direct Line session
func startBotSession() async throws -> String {
    let tokenResponse = try await authService.getDirectLineToken()
    UserDefaults.standard.set(tokenResponse.conversationId, forKey: "botConversationId")
    return tokenResponse.token
}

Microsoft documentation confirms: The Direct Line token is valid for 30 minutes. You can obtain a new token from the /v3/directline/tokens/refresh endpoint.

LUIS or CLU: Which to Choose for a New Project?

Bot Framework traditionally used LUIS for intent recognition. But Microsoft has migrated to CLU (Conversational Language Understanding) as part of Azure AI Language. CLU works twice as fast as LUIS, supports 50+ languages, and better recognizes complex dialogs with context. If you're starting a new project — choose CLU. For existing LUIS projects, prepare for migration: the export formats and SDKs differ (Azure.AI.Language.Conversations instead of Microsoft.Azure.CognitiveServices.Language.LUIS). We have helped clients migrate five projects in recent years.

Switching to CLU saves 30–40% of transaction costs and 40% of development time. Get a consultation from our engineer to evaluate the benefits for your project.

Storing Dialog State in Production

By default, Bot Framework stores UserState and ConversationState in memory. After a server restart, context is lost. For production, we use CosmosDbPartitionedStorage or BlobStorage. Configuration example in C#:

var storage = new CosmosDbPartitionedStorage(new CosmosDbPartitionedStorageOptions {
    CosmosDbEndpoint = configuration["CosmosDb:Endpoint"],
    AuthKey = configuration["CosmosDb:AuthKey"],
    DatabaseId = "BotStorage",
    ContainerId = "DialogState"
});

var userState = new UserState(storage);
var conversationState = new ConversationState(storage);

This ensures context preservation even when scaling to 1000+ simultaneous dialogs. If you have questions about setting up CosmosDB, contact us — we'll help.

Protocol Choice: WebSocket vs Polling

Direct Line supports two message retrieval modes: long polling (REST) and WebSocket. Let's compare key metrics.

WebSocket is preferable for active chat. On Android use OkHttp WebSocket, on iOS use URLSessionWebSocketTask. An important nuance: streamUrl lives about 60 seconds without activity, after which the connection closes. You need to handle onClosed and reconnect with a refreshed token.

Limitations of Adaptive Cards for Mobile UI

Adaptive Cards are a JSON schema for UI cards that Bot Framework uses for multi-channel rendering. There are official SDKs for iOS and Android (AdaptiveCards-iOS, adaptivecards-android), but style customization is limited: you cannot override fonts, margins, or animations. In 80% of our projects, we abandon Adaptive Cards in favor of custom event activities that render UI natively. This gives full control over design and behavior — for example, you can embed interactive forms or graphics.

What’s Included in a Turnkey Integration

• Architectural documentation (Direct Line scheme, dialog diagram)
• Repository with bot code and SDK for iOS/Android
• Access to Azure resources (Bot Service, CLU, CosmosDB)
• Deployment instructions via CI/CD and monitoring (Application Insights)
• Team training (2 hours online with typical error analysis)
• Compatibility guarantee with iOS 15+ and Android 12+

We have been working with Bot Framework v4 since its first release. We'll evaluate your project in one day — just write to us in the chat on the website.

Work Process

1. Analysis: audit current chat scenarios, gather requirements.
2. Design: choose stack (CLU/LUIS), design dialogs, Direct Line architecture.
3. Development: implement bot in C# (.NET) or TypeScript, create mobile client with WebSocket support.
4. Testing: Bot Framework Emulator for unit tests, load testing at 500+ RPS.
5. Deployment: Azure Web App with auto-scaling, setup monitoring and alerts.

Timeline Estimates

Integration with an existing Azure Bot — 3–5 days. Development from scratch (including CLU model, dialog logic, Azure infrastructure, and mobile client) — 2–4 weeks.

Request a consultation — we'll show you how to save up to 40% of your chatbot development budget.

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.