Move from provider-neutral model access to retrieval, tool-using agents, and Model Context Protocol. Each live tutorial is built around runnable C# code, explicit version baselines, and verification evidence so you can see what was actually tested.
The hub now covers four working layers and is expanding into evaluation and observability. Follow the sequence if you want a progressive path, or jump directly to the problem you need to solve.
Use IChatClient so application code can switch providers without rewriting the core.
Combine PostgreSQL full-text search, pgvector similarity, embeddings, and Reciprocal Rank Fusion.
Build an evidence-first Agent Framework workflow with read-only tools and explicit reasoning boundaries.
Build a real MCP server and client over stdio, discover tools, invoke them, and verify the protocol path.
Add repeatable quality checks, regression datasets, scoring, and evaluation gates before model or prompt changes ship.
Connect AI requests, retrieval, tool calls, latency, and failures to OpenTelemetry traces and operational dashboards.
Each guide below focuses on one architectural problem and links to a companion repository sample. The tutorial explains the design; the merged GitHub sample provides the concrete implementation and verification record.
Build an ASP.NET Core gateway around IChatClient and switch between Ollama, OpenAI, OpenRouter, and Azure OpenAI through configuration while keeping the application layer provider-neutral.
Combine PostgreSQL full-text search and pgvector semantic retrieval, generate embeddings through IEmbeddingGenerator, and fuse rankings with Reciprocal Rank Fusion.
Build a tool-using agent with AIAgent, AgentSession, streaming, three read-only tools, deterministic evidence, and explicit boundaries between facts, guidance, inference, unknowns, and unproven claims.
Build a real Model Context Protocol server with the C# SDK, expose three deterministic read-only tools over stdio, connect with a real MCP client, discover and invoke the tools, and verify the protocol path without an LLM or API key.
ModelContextProtocol 2.2.0 with .NET 10These two tutorials extend the current collection from building AI features into measuring and operating them. They are planned next and are not live yet.
Build a repeatable evaluation workflow for .NET AI features using curated test cases, measurable quality criteria, regression checks, and evidence you can compare across model or prompt changes.
Trace model requests, retrieval stages, tool calls, latency, failures, and cross-service activity so AI behavior can be investigated with the same operational discipline as the rest of a .NET system.
Pick the live tutorial that matches the boundary you are trying to solve. You do not need to complete the whole learning path before using one of these guides.
AI APIs move quickly. This collection avoids vague “works with the latest version” claims and instead records concrete package baselines, test results, and integration evidence for each companion sample.
source + pinned versions
│
▼
restore ──► Release build ──► deterministic tests
│
▼
real integration path
│
▼
GitHub companion
│
▼
CI checks
│
▼
tutorial documentationUse these primary references alongside the tutorials when you need authoritative API or protocol details. The dotnet-guide.com pages focus on tested implementation patterns and explain how the pieces fit together.
If your application still talks directly to one provider SDK everywhere, start with the gateway. If model access is already clean, choose retrieval, agent tools, or MCP based on the problem in front of you.