crestapps-core-chat

CrestApps.Core Chat Completions

This skill is the server-side completion and orchestration surface. Persisted conversations, hubs, chat UI and response handlers are covered by crestapps-core-chat-interactions; voice sessions by crestapps-core-realtime-voice; providers by crestapps-core-openai, crestapps-core-azure-openai, crestapps-core-ollama and crestapps-core-azure-ai-inference.

Guidelines

• CrestApps.Core exposes IAICompletionService (namespace CrestApps.Core.AI.Completions), not IChatCompletionService. It returns Microsoft.Extensions.AI ChatResponse and ChatResponseUpdate.
• Resolve the AIDeployment first, through a slot: ResolveSlotAsync(AIDeploymentSlotNames.Chat, ...) for answers and ResolveUtilityOrDefaultAsync(...) for background work. Do not construct provider clients in application code.
• Pass an AICompletionContext for the system message, sampling, tool selection, data source, and model parameters.
• Use IOrchestrator.ExecuteStreamingAsync(...) when the request needs a profile's tools, agents, RAG, MCP connections, or progressive tool scoping.
• The deployment's declared capabilities are enforced on every request. Tools, a JSON response format, reasoning options and streaming are removed or downgraded when the deployment does not declare toolCalling, structuredOutputs, reasoning or streaming.

Registration

using CrestApps.Core.AI.OpenAI;

builder.Services.AddCrestAppsCore(crestApps => crestApps
    .AddAISuite(ai => ai
        .AddOpenAI()
    )
);

Configure a connection and at least one deployment that declares textGeneration (plus toolCalling and streaming for chat). See the provider skills for the configuration shape.

Stream a Completion

using System.Runtime.CompilerServices;
using CrestApps.Core.AI.Completions;
using CrestApps.Core.AI.Deployments;
using CrestApps.Core.AI.Models;
using Microsoft.Extensions.AI;

namespace MyApp.Services;

public sealed class SupportAnswerService
{
    private readonly IAICompletionService _completionService;
    private readonly IAIDeploymentManager _deploymentManager;

    public SupportAnswerService(
        IAICompletionService completionService,
        IAIDeploymentManager deploymentManager)
    {
        _completionService = completionService;
        _deploymentManager = deploymentManager;
    }

    public async IAsyncEnumerable<string> StreamAsync(
        string prompt,
        string deploymentName = null,
        [EnumeratorCancellation] CancellationToken cancellationToken = default)
    {
        var deployment = await _deploymentManager.ResolveSlotAsync(
            AIDeploymentSlotNames.Chat,
            deploymentName,
            cancellationToken: cancellationToken)
            ?? throw new InvalidOperationException("No deployment can fill the chat slot.");

        var context = new AICompletionContext
        {
            SystemMessage = "You are a concise support assistant.",
            Temperature = 0.2f,
            MaxTokens = 800,
        };

        // Applied only when the deployment declares the reasoning feature and the reasoningEffort parameter.
        context.ModelParameters[AIDeploymentParameterNames.ReasoningEffort] = "Low";

        await foreach (var update in _completionService.CompleteStreamingAsync(
            deployment,
            [new ChatMessage(ChatRole.User, prompt)],
            context,
            cancellationToken))
        {
            if (!string.IsNullOrEmpty(update.Text))
            {
                yield return update.Text;
            }
        }
    }
}