AgenticFramework

A progressive tutorial demonstrating AI agent patterns in C# — from basic chat to multi-agent orchestration, RAG, observability, and a PostgreSQL DBA agent.

Each numbered project builds on the previous one. All .cs files are self-contained top-level programs using #:package directives — no .csproj or .sln files needed.

Projects

#	Project	Description
01	ConsoleChat	Basic streaming console chat with OpenAI
02	WebChat	Stateless web API chat with tool support (ASP.NET Minimal API)
03	AgentLoop	Autonomous agent loop: call model → execute tools → feed results → repeat
04	HumanInTheLoop	Agent loop gated by user approval before tool execution
05	WebHumanInTheLoop	Web agent with dual execution (JS + C# Roslyn), skill persistence, approval UI
06	AgentToAgent	Multi-agent orchestration via A2A protocol (JSON-RPC 2.0) with SSE streaming
07	RagObservability	RAG with Qdrant vector DB + full OpenTelemetry instrumentation + Grafana
08	A2UI	Agent-to-UI dashboard: agent calls tools, frontend renders live widgets
09	PgDba	PostgreSQL DBA agent with monitoring, diagnostics, and approval-gated operations

Quick Start

Prerequisites

• .NET 9+
• Docker & Docker Compose
• OPENAI_API_KEY environment variable

Run any project

export OPENAI_API_KEY=sk-...

# Console chat
dotnet run --file 01_ConsoleChat/Chat.cs

# Web projects (open http://localhost:5000)
dotnet run --file 02_WebChat/Api.cs

Database setup (for project 09)

./pipeline.sh              # Start PostgreSQL on port 5488, load ~15GB pgbench data
./pipeline.sh small        # ~7GB variant
./teardown.sh              # Stop and clean up

RAG + Observability stack (for project 07)

cd 07_RagObservability && docker compose up -d && cd ..
dotnet run --file 07_RagObservability/Api.cs
# App: http://localhost:5000 | Grafana: http://localhost:3000 | Qdrant: http://localhost:6333

React frontends (projects 08, 09)

cd 08_A2UI/app && npm install && npm run dev    # http://localhost:5173
cd 09_PgDba/app && npm install && npm run dev   # http://localhost:5173

Problem Simulation Scripts

The problems/ directory contains scripts that simulate real PostgreSQL performance issues:

Script	Issue
01-slow-queries.sh	Sequential scans, high query duration
02-lock-contention.sh	Row-level locks, waiter pileup
03-connection-exhaustion.sh	Max connections reached
04-table-bloat.sh	Dead tuples accumulation
05-cache-miss.sh	Low buffer cache hit ratio
06-temp-files.sh	Disk spillover, high I/O
07-xid-wraparound.sh	Transaction ID approaching limit
08-idle-in-transaction.sh	Long-held idle transactions
09-checkpoint-spikes.sh	Heavy checkpoint load
10-unused-indexes.sh	Indexes with zero scans

source problems/common.sh
bash problems/01-slow-queries.sh
bash problems/monitor.sh          # live dashboard
bash problems/reset.sh            # clean up

Key Patterns

Tool Definition

[Description("Fetches content from a URL")]
static async Task<string> Curl([Description("target URL")] string url)
    => await new HttpClient().GetStringAsync(url);

var tools = new[] { AIFunctionFactory.Create(Curl) };

Agent Loop

while (true) {
    var response = await chatClient.GetResponseAsync(messages, new ChatOptions { Tools = tools });
    messages.AddRange(response.Messages);

    var toolCalls = response.Messages
        .SelectMany(m => m.Contents.OfType<FunctionCallContent>()).ToList();
    if (toolCalls.Count == 0) break;

    foreach (var call in toolCalls) {
        var tool = tools.First(t => t.Name == call.Name);
        var result = await tool.InvokeAsync(new AIFunctionArguments(call.Arguments!));
        messages.Add(new(ChatRole.Tool, [new FunctionResultContent(call.CallId, result)]));
    }
}

Libraries

• Microsoft.Extensions.AI — Chat client abstraction
• Microsoft.Agents.AI.OpenAI — Agent framework with approval gates
• Microsoft.Agents.AI.Hosting.A2A — Agent-to-Agent protocol
• OpenAI SDK v2.9.1 — Direct API client
• Qdrant.Client — Vector database for RAG
• Npgsql — PostgreSQL driver
• OpenTelemetry — Traces, metrics, instrumentation
• Microsoft.CodeAnalysis.CSharp.Scripting — Roslyn eval (project 05)
• CopilotKit — React dashboard widgets (projects 08, 09)

Environment Variables

Variable	Required	Used by
OPENAI_API_KEY	Yes	All projects
GITHUB_TOKEN	No	07 (GitHub repo indexing)
PG_CONNECTION	No	09 (default: localhost:5488)

Learning Path

1. 01–02 — Chat basics + tool calling
2. 03–04 — Agent loops + human approval
3. 05 — Skill persistence + dual execution
4. 06 — Multi-agent orchestration
5. 07 — RAG + observability
6. 08–09 — Dashboard rendering + domain specialization

License

MIT