Your own AI agent: how to build memory the model reaches for itself

Mandatory registration and purchase of tickets: https://stacja.it/produkt/twoj-wlasny-agent-ai-jak-zbudowac-pamiec-po-ktora-model-siega-sam/

Weekend workshop at 9:00-13:00 (4 h)

What will you learn? - You will learn three types of memory of language models: short-term, long-term and external context - You will understand how the document database differs from memory and why just adding files to the chat does not create permanent memory. - You will see how the agent working directly on the files completes the missing information writing path. - You will learn the document card, thanks to which the model recognizes the type of document, its timeliness and scope. - You will learn to build a material map that allows the model to independently select the information needed instead of receiving the whole context at once. - You will see in practice how the prepared system allows the model to enter into the topic without additional introduction and without asking about the basics. - You will learn the ritual of closing the conversation, in which the agent saves new arrangements and updates information that has ceased to be true. - You will learn how to determine the rhythm of maintaining the knowledge base and the rules for removing outdated information from it. - You will receive a ready-made document card template and a description of the ritual, which you can immediately use at home.

Who is the workshop for? For people who work with an AI assistant every day and return to the same topics for weeks: consultants, strategy and marketing people, project managers, analysts, small business owners. You don’t need to be able to program. The tools we use were created for programmers, but here they are used only to work on plain text documents.

Prerequisites - Daily work with the language model (ChatGPT, Claude or Gemini). - Without programming knowledge and without materials prepared in advance (everything is created on the workshop). - Openness to work with text files instead of the chat window itself.

Installation requirements: - Account in payment plan Claude, ChatGPT or Gemini. The free plan does not give you access to a tool that works on files. - Installed tool reading and saving files on disk: Claude Code, Claude Cowork, Codex or Antigravity. Markdown: Obsidian or Visual Studio Code, both free.

Marek Kibała: AI trainer and implementation practices. He conducts multi-day training on the implementation of AI in companies and institutions, in several cities, for hundreds of participants, from rooms that open ChatGPT for the first time, to teams working with tools for months.

In addition to training, it builds AI systems running on organizational documents: knowledge bases that correspond with the indication of a specific place in the document, and automations in which the last word always belongs to a person. He sits down to each tool with the assumption that he is wrong somewhere, and he teaches this search. A participant who knows the limits of the tool still uses it six months after training.

The one who is promised reliability puts it off after the first mistake.

Program 1. How the model remembers: short-term memory, long-term memory and external context. Why is the database only readable and the memory must be saved 2. Agent working on files: a tool instead of a chat window.

Where does it read, how does it type, and how does it differ from uploading files to conversation 3. Document tab: a header by which the model recognizes the type, timeliness, and boundary of the document 4. Material map: the list which model reads first and chooses from it what it needs.

Test: A fresh conversation enters the topic without asking 5. Ritual at the end of the conversation: a diary to which the agent himself adds findings and what has ceased to be true. Maintenance and cleaning of the base 6. Q&A

The number of places is limited! The organizer reserves the right to cancel the event in case of failure to register the minimum number of participants.

Mandatory registration and purchase of tickets: https://stacja.it/produkt/twoj-wlasny-agent-ai-jak-zbudowac-pamiec-po-ktora-model-siega-sam/