Keeping
the Why

Why you might want it

Knowing why you rejected an option can be as useful as knowing what you chose.

Imagine you're planning a website where people can book local workshops. You decide people should be able to book without creating an account, because you want a first booking to be easy. You set accounts aside for now, with a reason to reconsider: repeat customers asking for a way to manage their bookings.

Months later, the record helps you see why you made that choice and whether it's worth reopening. A fresh assistant can work from the reasoning instead of asking you to reconstruct it.

That continuity matters when you return after time away, switch AI tools, or bring another person into the project. You can ask what was decided, what is still open, and why.

Persimyn suits work with enough duration or complexity that you expect to return to it, change direction, explain it, or hand it off. That might be a product team or one person developing an idea over several months. It is designed for work where mistakes are recoverable; it isn't built for legal, regulated, or safety-critical work where signing off on every word is part of the process.

From working notes to something you can share

A short project brief might explain a choice in one paragraph. Getting there could have taken several conversations, research, and ideas that didn't work out. Persimyn keeps those details as source material for your assistant. The same record can support further work or help you prepare a product requirements document, presentation, or executive summary.

You don't need to read every note. You discuss the ideas and confirm your choices in conversation. When you need something to share, ask your assistant to draw on the record and shape it for that reader. It's easier to turn a page of detail into a paragraph than recover reasoning that was never written down.

Your first project

Start with something you actually want to work through. One question is enough; you don't need a complete project description or documents prepared in advance.

Download the starter kit. I can help you get set up while Persimyn is in early access. You'll need an AI assistant that can read and write files in a project folder. Claude Code and ChatGPT Work are the setups tested so far. The kit is free; access to your AI tool is separate.

Set up your project folder

For a new project:

1. Unzip the starter kit.
2. Rename the folder containing AGENTS.md to your project's name and put it wherever you keep your work.
3. Open that folder in your AI assistant. Keep AGENTS.md and CLAUDE.md in place: they provide the instructions your assistant reads.

If you have notes, research, or exports to bring into this new project, copy them into docs/raw/ and tell the assistant they're there. Keep their folder structure intact. These copies stay as source material; adding them doesn't turn their contents into decisions you've adopted.

Already have a project folder? Keep working there. Add the kit's files and merge its folders without replacing existing content. If you already have a README.md, AGENTS.md, or CLAUDE.md, keep the kit's copies separately and ask your assistant to merge the relevant content, or ask me to help. Existing code and working files stay where they are; docs/raw/ is for source material you're bringing in, not the whole project.

The kit's START-HERE.md has setup instructions for the version you receive.

Have a conversation

The project instructions load automatically in a supported setup. Start with what you want to work on. For example:

Talk through the question as you normally would. Ask for alternatives, challenge an assumption, explain a constraint. The assistant should offer to keep things worth returning to. You can also ask it to save something yourself.

An idea can stay a working note. Questions can stay open, with a note about what would answer them. When you make a choice, the assistant summarizes what it will record: what you chose, what you set aside, and what would make you reconsider. Correct anything it misunderstood, then confirm when the summary reflects your choice. You can also say no or keep talking.

After discussing the options, you might say:

The assistant should offer to record the choice and its reasoning. You can simply say “Yes, record that” once its summary is right. It then writes the decision into your project and shows you what it recorded, attributed to the person who made the choice. If you agree on what to do next, it can keep that plan too.

See what you can return to

Once you've recorded a useful choice, open a fresh conversation in the same project and ask:

The answer should recover the reasoning and point to the saved decision. For this example, it might say:

You've worked through a question and left something the next conversation can build on. Another assistant with access to the same files can use that record too.

If something doesn't work: correct misunderstandings as you would in any conversation. If the assistant never offers to record anything, or can't find a choice you saved, let me know so I can help check the setup. You shouldn't need to remind it to read the project instructions.

Where the notes go

The assistant handles the filing. Each kind of material has its own place:

your-project/
├── AGENTS.md          the working rules your assistant reads
├── CLAUDE.md          points Claude to those rules
├── README.md          what the project is, for anyone arriving
├── NEXT_STEPS.md      the plan you've agreed, added when needed
└── docs/
    ├── explorations/  ideas and working notes, grouped by person
    ├── decisions/     choices you've made, and why
    ├── foundation/    reference notes about how things work
    └── raw/           source material, kept as it arrived

Decisions are kept by topic, with their reasoning and alternatives together. You can change your mind later; the earlier reasoning stays to explain what changed.

These are ordinary Markdown files in ordinary folders. You can inspect them, keep them with your code or other project material, and take them with you if the tools around them change.

Working with others

You can use Persimyn on your own, with different AI assistants, or with other people.

Everyone needs access to the same project files, and their assistants need to be able to read and update them. That shared space could be a GitHub repository or a shared folder through a service such as Google Drive or Dropbox, depending on your tools.

Each person keeps working notes in their own named folder under docs/explorations/. Decisions become part of the shared record, attributed to the people who made them. You arrange sharing, access, and how changes reach each other through the platform you use. Shared setups are still being tested during early access.

Early access

Persimyn is still being tested. Bring a project you care about and a question you want to work through. I can help you get started and work through anything confusing, including working with others.

Your experience will help improve both the kit and these instructions. Tell me where the assistant keeps useful notes, where you have to step in, and whether returning to the project is easier. If something feels like more effort than it's worth, I'd like to hear that too.