Skip to content

README.md

Latest commit

 

History

History
127 lines (94 loc) · 6.22 KB

README.md

File metadata and controls

127 lines (94 loc) · 6.22 KB
name .NET MAUI - Local Chat Client with Tools
description Chat UI demonstrating on-device Apple Intelligence function calling (tools) with weather, calculator, file listing, system info, and timers — no API keys or cloud services required.
page_type sample
languages
csharp
xaml
products
dotnet-maui
urlFragment local-chat-client-with-tools

LocalChatClientWithTools (MAUI + On-Device Apple Intelligence Tools)

A .NET MAUI sample showing how to enhance an on-device LLM with Microsoft.Extensions.AI function calling (tools) using Apple Intelligence. All inference runs locally — no API keys, no cloud services, no cost.

Prompt buttons Weather tool Calculator tool Multi-tool calls
prompts weather calculator multi-tool

What you'll learn

  • Using AppleIntelligenceChatClient from Microsoft.Maui.Essentials.AI as a local IChatClient
  • Supplying a set of AIFunction tools to IChatClient via ChatOptions.Tools
  • Implementing strongly described tool schemas (JSON schema) for better argument selection
  • Resolving user-friendly locations to coordinates with the free open-meteo.com API
  • Building a MAUI chat UI with template selectors for different message types (text, tool calls)
  • Running AI inference entirely on-device with no external API keys

Prerequisites

  • .NET 10 SDK
  • iOS 26.0+ or macCatalyst 26.0+ device/simulator with Apple Intelligence enabled
  • No API keys or cloud accounts needed

Tools implemented

Tool Name Purpose Notes
WeatherTool get_weather Current conditions for a location Geocodes via open-meteo.com, then fetches forecast — free, no API key
CalculatorTool calculate Evaluate arithmetic / percentages Sanitizes expression & returns formatted result
FileOperationsTool list_files List files & folders in a common or given path Limits count; resolves shortcuts (Documents, Desktop, Downloads)
SystemInfoTool get_system_info Battery, storage, memory, device info Returns live device values on Apple platforms
TimerTool set_timer Create a one‑shot timer with title Keeps in-memory timers; shows alert on completion

Each tool is created via AIFunctionFactory.Create using strongly-typed methods and AOT-compatible JSON serialization via ToolJsonContext.

Key files

  • MauiProgram.cs – Registers IChatClient via AppleIntelligenceChatClient with UseFunctionInvocation() for automatic tool dispatch, and registers all tool functions.
  • Services/Tools/*.cs – Tool implementations using AIFunctionFactory.Create for clean, factory-based tool registration.
  • ViewModels/ChatViewModel.cs – Collects tools and invokes _chatClient.GetStreamingResponseAsync passing ChatOptions.Tools.
  • ViewModels/ChatMessageViewModel.cs – Base view model for chat messages, with TextMessageViewModel and ToolCallMessageViewModel subclasses.
  • MainPage.xaml – Chat layout with ChatTemplateSelector, EmptyView sample prompts, and a Clear toolbar item (resets to EmptyView).

Sample prompts

Try:

  • "What's the weather in Seattle?"
  • "Calculate 15% tip on $47.50"
  • "List the files in my Documents folder"
  • "Show me current battery level and available storage"
  • "Set a 5-minute timer for my coffee break"
  • "What's 25 × 18 + 150?"
  • "Check the weather in Tokyo and Cape Town"

Run

# iOS simulator
dotnet build -f net10.0-ios -t:Run

# macCatalyst
dotnet build -f net10.0-maccatalyst -t:Run

Use a sample prompt or type your own. Use the toolbar Clear to reset and revisit the instructional prompt list.

Note: This sample only runs on Apple platforms (iOS / macCatalyst) with Apple Intelligence. Other platforms will throw PlatformNotSupportedException at startup.

Architecture highlights

Tool invocation flow

  1. User sends natural language input.
  2. ChatViewModel builds ChatOptions with the registered tool objects.
  3. AppleIntelligenceChatClient (wrapped with UseFunctionInvocation()) decides if any tool(s) should be invoked and executes them, passing JSON arguments that conform to each tool's declared schema.
  4. Tool result objects surface in the assistant response and are bound to specialized UI cards.
  5. UI shows structured cards or plain text depending on ChatMessage flags.

On-device vs. cloud

This sample is a local-only variant of the ChatClientWithTools sample. The key differences:

ChatClientWithTools LocalChatClientWithTools
LLM provider Azure OpenAI (cloud) Apple Intelligence (on-device)
API keys Required (Azure + optional OpenWeatherMap) None
Weather API OpenWeatherMap (key required) open-meteo.com (free, keyless)
Platforms Windows, iOS, macCatalyst, Android iOS, macCatalyst only
NuGet packages Azure.AI.OpenAI, Microsoft.Extensions.AI.OpenAI Microsoft.Maui.Essentials.AI

UI features

  • EmptyView with categorized starter buttons (Weather / Calculator / Files / System Info / Timers).
  • Rich result cards styled with color-coded borders.
  • Toolbar "Clear" to wipe conversation state quickly (teaching scenario friendly).
  • Auto-scroll to the latest message.

Useful references

Notes

  • Requires iOS 26.0+ or macCatalyst 26.0+ with Apple Intelligence available and enabled.
  • All AI inference runs on-device — no data leaves the device for LLM processing.
  • Weather data comes from open-meteo.com (free, open-source weather API).
  • Timers are in-memory only (lost on app exit) and surface completion via a DisplayAlertAsync dialog.
  • File operations are intentionally conservative; expand with care for security.
  • This sample emphasizes pedagogy over exhaustive production hardening.

For the cloud-hosted version of this sample, see ChatClientWithTools in the same folder tree.