Why should you care?

Private company knowledge must be retrieved, not memorized by the base model.

See it live — copy this example

Run CLI steps in a terminal. Run C# samples in a .NET 8 console or Web API project with Ollama running on localhost:11434.

// Using pgvector — RAG sketch
var chunks = await retriever.SearchAsync(question, topK: 4);
var context = string.Join("\n---\n", chunks.Select(c => c.Text));
var prompt = $"""
Use ONLY this context. If missing, say you do not know.
CONTEXT:
{context}
QUESTION: {question}
""";
return await ai.CompleteAsync(prompt);

What happened?

• Search returns top chunks.
• The prompt includes only that context.
• The model is told not to invent.

Practice next

1. Read the example and rewrite it in your own repo.
2. Run it against local Ollama (or finish the Ollama module first).
3. Change one prompt constraint and compare output.
4. Tighten the prompt to 3 bullets max.
5. Log duration of the model call.

Remember

You can explain Using pgvector simply. You have a runnable local sketch. You know one risk to watch.

Using pgvector in LocalAIDesk

Your team applies using pgvector while building a private agent desk.

Outcome: A concrete next step exists in code or checklist form.