AI Tools

The AI conversation service supports Microsoft.Extensions.AI tools (AITool) that are included in every chat request. This lets the AI call functions on the user’s behalf.

How Tools Are Provided

Tools are supplied through the IContextProvider visitor pattern. Each registered IContextProvider populates an AiContext via its Apply() method, adding tools to the context’s Tools list. The AiContext also carries system prompts, quiet words, and speech options — so providers can influence the full request pipeline. The built-in ContextProvider automatically passes through any AITool instances registered in DI — so any library that registers AITool instances in your service collection will have those tools available to the AI via the default provider.

You can also supply tools from a custom IContextProvider:

public class MyToolProvider : IContextProvider
{
    public Task Apply(AiContext context)
    {
        context.Tools.Add(AIFunctionFactory.Create((string city) =>
        {
            return $"The weather in {city} is sunny, 72°F.";
        }, "GetWeather", "Gets the current weather for a city"));

        return Task.CompletedTask;
    }
}

builder.Services.AddSingleton<IContextProvider, MyToolProvider>();

Or register AITool instances directly in DI (picked up by ContextProvider):

// These tools are all automatically discovered and included in AI chat requests

// Shiny.DI source-generated tools
builder.Services.AddGeneratedAITools();

// Shiny.DocumentDb tools
builder.Services.AddDocumentStoreAITools(tools => { /* ... */ });

// Your own custom tools
builder.Services.AddSingleton<AITool>(sp =>
{
    return AIFunctionFactory.Create((string city) =>
    {
        return $"The weather in {city} is sunny, 72°F.";
    }, "GetWeather", "Gets the current weather for a city");
});

Built-in: Chat History Lookup

When an IMessageStore is registered, the built-in ContextProvider automatically adds a ChatLookupAITool to every request:

opts.SetMessageStore<MyMessageStore>();

This gives the AI the ability to search past conversations. Users can ask questions like:

• “What did we discuss yesterday?”
• “Find the conversation where we talked about deployment”
• “What was the recipe you gave me last week?”

The tool exposes these search parameters to the AI:

Parameter	Type	Description
messageContains	string?	Text to search for in messages
fromDate	DateTimeOffset?	Start of date range
toDate	DateTimeOffset?	End of date range
limit	int?	Maximum results to return

Custom Tools

You can register any number of custom tools. They’re standard Microsoft.Extensions.AI tools:

// Function-based tool
builder.Services.AddSingleton<AITool>(sp =>
    AIFunctionFactory.Create(
        async (string query, CancellationToken ct) =>
        {
            var db = sp.GetRequiredService<IDatabase>();
            var results = await db.Search(query, ct);
            return string.Join("\n", results);
        },
        "SearchDatabase",
        "Searches the application database"
    )
);

Note

All registered IContextProvider implementations are executed in sequence, each adding their tools to the shared AiContext. The aggregated tools are then included in the chat request.