AiCleverness 1.9.0

AiCleverness — Provider-Neutral AI Execution Runtime for .NET

AiCleverness is a lightweight execution runtime for building AI-powered .NET applications. Agents are one execution pattern; the core value is orchestration: policies, planning, deterministic strategies, tool execution, quality gates, transformations, and observability around any provider adapter.

Read the developer manual — getting started guides, core concepts, tool authoring, streaming, memory tiers, security, workflows, and a full API reference.

Installation

dotnet add package AiCleverness

Requires .NET 10.0 or later. Zero external AI provider SDKs — bring your own.

What You Can Build

Quick Start

using AiCleverness.Abstractions;
using AiCleverness.Models;

var services = new ServiceCollection();
services.AddAiClevernessRuntime();
services.AddAiClevernessLlmClient<MyLlmClient>();   // your provider adapter
services.AddAgentTool<WeatherTool>();                // your tool

var provider = services.BuildServiceProvider();
var runtime = provider.GetRequiredService<IAgentRuntime>();

var result = await runtime.RunAsync(new AgentRequest(
    Goal: "What is the weather in Tokyo?",
    AllowedToolNames: ["get_weather"]));

Console.WriteLine(result.Output);

Three pieces — ILlmClient, ITool, DI wiring — and you have a working execution runtime.

Transcripts and decision-tree diagnostics

Transcript persistence is opt-in. For ordinary agent runs, configure an absolute directory in the request and provide a redactor in AgentRuntimeOptions:

services.AddAiClevernessRuntime(options =>
{
    options.TranscriptRedactor = text =>
        text.Replace("customer@example.invalid", "[REDACTED]", StringComparison.OrdinalIgnoreCase);
});

var request = new AgentRequest(
    Goal: "Summarize the evidence",
    Parameters: new Dictionary<string, object>
    {
        [AgentPropertyKeys.MarkdownTranscriptDirectory] =
            Path.GetFullPath("transcripts")
    });

Normal transcripts pass persisted content through the host redactor. Debug transcripts are explicitly unredacted and should be used only with controlled data and access. The built-in Markdown builder and file sink work without additional setup once the directory and redactor are configured.

Decision-tree applications configure DecisionTreeExecutionOptions with an absolute TranscriptDirectory, optional DecisionTranscriptPolicy limits, and the same redaction/debug choices. Both agent and decision-tree options support per-execution TranscriptBuilderFactory and TranscriptSinkFactory delegates for JSON/HTML/structured formatting or non-file destinations such as databases and queues. For small Markdown changes, derive from TranscriptBuilderDecorator and override only the section you need; implement ITranscriptBuilder directly only for a completely different representation. Factories must create fresh builder and sink instances; never return cached mutable transcript objects or register them as singletons. Transcript failures are best effort and do not replace the primary execution result. See the full decision-transcript guide for action outcome summaries, readable node names, policy limits, logical sink paths, and failure semantics.

Full Reference

The repository README contains the complete reference: core concepts, full DI setup, models, streaming, memory tiers, security, workflows, persistence, observability, and agent scoping — github.com/AlexNek/AICleverness.

• Developer manual
• API reference
• Changelog

License

MIT