1. Why projects exist at all
A chat ends, and the work on one client is scattered across mail, SharePoint, notes and memory. Ask a fresh chat "Where are we on Acme?" and it asks you to retell the whole account. Inside a project, the brain opens what it already knows and answers.
A project is the unit of memory for any piece of work that lasts longer than one conversation: a client mandate, a search, a pilot.
2. The four memories (WESP)
- Working: the chat, what you are saying right now. It lives inside one conversation.
- Episodic: the diary, what happened and when. One entry per day, written as you work.
- Semantic: the wiki, what things are: people, companies, projects. Projects live here.
- Procedural: the skills, how things are done: mail, candidate sheets, reports. Skills also connect the brain to external sources.
The working memory looks things up in the diary and the wiki, then acts through a skill. A skill is also how the brain reaches HireSweet, LinkedIn, SharePoint, the web and email.
3. A notebook, not a library
Your diary is a notebook: notes, decisions, what happened, where things are. It stays small (a year of daily use is under 30 MB) and quick to re-read.
The library is your sources: SharePoint, HireSweet, LinkedIn, mail, the web. Gigabytes of documents stay where they are, and the brain knows where to look. You can upload a document and the brain will read it and take notes, but do not expect the brain to become the archive.
4. What a project is
A text file that points to everything about a piece of work: the diary entries about it, the wiki pages of the people and companies involved, the skills it needs, the mail threads and the external resources. It is organised to keep everything that matters, complete and short, readable by the brain and by you in half a minute.
5. What is inside
- A description, key info and any KPIs.
- Do and Don't: how to work on this project.
- Skills: the ones this project needs, loaded when you open it.
- Current state: one line, today's.
- Anything meaningful to keep it consistent. Keep the page short and readable: the brain explores the external sources for the details.
Acme Logistics: Warehouse Manager search What it is. We are finding a Warehouse Operations Manager for Acme, a logistics client in Antwerp. Contact: Maria Dupont (HR). Role open since 22 Sep, shortlist by 9 Oct, 3 candidates interviewed. Do answer the client the same day. Don't send a CV without the candidate's OK. Skills. candidate-presentation, HireSweet search. Current state. Shortlist of 3 sent, client interviews being scheduled. Resources. HireSweet project "Warehouse Manager Acme" - LinkedIn profiles of the 3 candidates - job description on SharePoint.
6. Current state, and the whole story
The project carries the current state: one line, replaced when things change. It does not carry the history. The history lives in the diary: every entry about the client is linked to the project, so the whole story can be rebuilt by reading those entries in order. The diary is the film; the project is today's photo.
7. Project status
- Experiment: an idea or a test. It may die, and that is fine.
- Beta: it runs on real cases and a person checks every result.
- Active: trusted, part of the daily work.
- On hold: paused, nobody is waiting. It can reopen.
- Archive: closed, no more work on it.
You choose the stage, or ask the brain: "put Acme on hold". It is a label on your own project and nobody else reads it.
8. A private notebook, and how to share
The project is private to you and your brain. To share, there are email, the Village (shared channels, tasks and decisions for the team) and artifacts (pages, sheets and reports for clients and candidates). These are the deliverables, and more are coming. The Village is the next session, 20 October; artifacts follow on 27 October.
9. Activating a project
Type /project <name> and the brain loads everything: the summary and the phase, do and don't and the current state, the skills the project needs if it has specific ones, and what it links to (diary entries, wiki pages, mail threads, people). Short page first, details only when needed.
10. Target: one chat, one target
Open the chat with the project (existing or new) and what the chat is for. Example: /project Acme, target of this chat: answer the open mails from Maria and put the status report on SharePoint. Work until it is done; when the target is reached, start a new chat: fresh context, a shorter chat, easier to find later. One chat for a whole project works too; one chat for one target is the best practice.
11. How the brain answers, and with a skill on the fly
Ask "Where are we on Acme?" and the brain opens the project (current state), reads the latest diary entries, checks the latest mail and goes to the source if needed, then answers without asking you to retell anything.
Add a skill in the same request: /project Acme generate a report on it with /artifact. The brain does the same lookups, loads the artifact skill and builds the report as a page you can share.
12. When the brain does not do what you want
Correct the answer if you must, but fix the source so it holds next time. Three places can hold a correction:
- The project (most often): for this project only. "For Acme, always write the sheet in French."
- The skill (sometimes): a way of working that applies in every project.
- The boot, the brain's base rules (very rarely): for everything, handle with care.
Ask the brain to record the correction where the rule belongs. This is a bigger topic that we take up later in the course.
13. Your first project, and the check
- Ten minutes, one real client.
- Type
/project <client>; the brain interviews you, a few questions at a time, then shows a draft and writes the project when you approve. - Already have a project? Do not start over: open it and ask "What is missing?" The brain reads it and asks about the gaps.
Check list for any project: a short summary, key info (client, contacts, dates, KPIs), do and don't, a phase, a one-line current state, and a history made of linked diary entries.
14. Where grbrain is going
From David, Lorraine and the sessions so far: value you can use immediately, adoption starting from tasks you already do, branded pages first (a candidate presentation from a CV and a job description), and worked examples that are short, written and ready to copy.
The pilot asked by Polina has three outputs: client brief and feasibility, candidate presentation and mission progress report, built around one reference record per mission. That record is the project. For now grbrain is a narrower brain: candidates first, specific skills, simple pain points.
Glossary
- Project: the page that gathers everything about one piece of work.
- Diary: dated entries about what happened; the history of a project.
- Wiki: pages about people, companies and projects.
- Skill: a way of doing something, such as writing a candidate sheet or connecting to a source.
- Artifact: a page, sheet or report generated by the brain to share.
- Target: what one chat is for.
- WESP: Working, Episodic, Semantic, Procedural: the four memories.
Further reading
- How AI Agents Remember Things (video, English). how one AI agent framework (OpenClaw) organises memory, with a mention of Claude's own memory feature. A general introduction, not specific to our brain.