Skip to content

AI Basics

Build an AI setup
on your own machine.

A series for developers who want to work with language models without sending their code and data anywhere. We start with a bare machine and finish with an agent that knows your project — free of charge, and without an account in any cloud.

The episodes themselves are written in Polish. The code, commands and diagrams are language-independent, so they are usable either way.

  1. 01Uruchamianie lokalnych agentów AI w oparciu o OllamaJak uruchomić modele językowe na własnym komputerze: instalacja Ollamy, wybór modelu, lokalne REST API i podpięcie pod edytor kodu.Verified: September 2026
  2. 02Budowa agenta z dostępem do bazy wiedzy projektu (RAG) w oparciu o OllamaJak dać lokalnemu modelowi dostęp do kodu i dokumentacji projektu: indeksowanie repozytorium, embeddingi, wyszukiwanie i ocena jakości odpowiedzi.Verified: September 2026
  3. 03Pamięć długoterminowa agenta AI w kontekście projektu (Ollama)Jak sprawić, żeby agent pamiętał ustalenia z poprzednich rozmów: wyciąganie faktów, ocena istotności i trwałe przechowywanie wiedzy o projekcie.Verified: September 2026
  4. 04Kompletny agent programisty: łączymy Ollama, narzędzia, RAG i pamięćSpinamy RAG, pamięć długoterminową i wywoływanie narzędzi w jedną pętlę decyzyjną — działającego lokalnie agenta programisty opartego o Ollamę.Verified: September 2026

In progress

The first four episodes close out the local setup. The next five take the same agent further — off the laptop, into the cloud, and into the hands of the people who will use it. Each one is published here only after every step has been walked through live, exactly like the ones before it.

  • 05From a local setup to Azure AI Foundry — what changes, and whyThe same agent loop, a different environment. An endpoint instead of localhost, an identity instead of a key, a price per token instead of spare RAM — and a data boundary that finally matters.
  • 06Design the application locally, run it in the cloudOllama as a free workshop: prompts and the agent loop are refined on your own machine, and only the finished application — an assistant for employees — goes out to Foundry.
  • 07The same agent as a Microsoft Teams add-inThe assistant arrives where the team already works: a bot in a private chat, sign-in through Entra ID, tenant admin consent, and the things that usually break on a first rollout.
  • 08Azure AI Search — a company knowledge base instead of local RAGThe RAG from episode 2, scaled to a whole organisation: the index, hybrid search, and the hardest part — permissions for documents the person asking should not see.
  • 09Per-user memory, or why the answer depends on who is askingThe memory from episode 3, attached to a person rather than a project. A key from Entra ID, the rule that memory scope equals conversation scope, and the line between private memory and team knowledge.