S1: Build your own information system

English edition of S1. The remaining skills (S2–S9) are published in Japanese; they are being translated.

Before AI defines you, define yourself — first, and by yourself.

What you get from this skill

A state in which your values, judgment criteria, and ways of thinking are accumulated and managed on your own PC — a setup where you can use AI to strengthen your own thinking rather than replace it.

Why it matters

Even when you believe you are thinking for yourself, you are often just moving the way AI suggested. You receive the recommended information, you pick the proposed answer — and as this piles up, the line between “my own thought” and “the AI’s suggestion” disappears.

Holding your own data yourself is the starting point of everything.

The method: building an Ontology system

S1 is not a skill about “memorizing something.” It is the skill of building, on your own PC, a system that records, analyzes, and continually updates your own thinking.

The core idea: make it a white box

Most AI apps are black boxes — you cannot see what is happening inside. S1 aims for the opposite: a state where your data’s location, and what the AI is analyzing and how, are fully visible. All data lives in plain files on your own machine. Even if a cloud service shuts down, your data stays in your hands.

Step 1: prepare your tools

Before making folders, get two kinds of tools ready.

① An IDE (an app to edit and manage files)

Used to create and edit Markdown (.md) files.

App	Notes
Cursor	An IDE with a free plan; easy to pair with Claude Code
Antigravity	Google’s IDE, with Gemini built in from the start

You do not have to write the Markdown yourself. Ask the AI: “make me a concept file on this theme,” and it will write it.

② An AI agent (to create and analyze the content of files)

AI tool	Notes
Claude Code	Strong at file analysis, updating the Ontology, generating reports
Gemini (built into Antigravity)	Comes bundled with the IDE under one Google One plan — the low-cost option

Either is fine. Antigravity bundles the IDE and the AI (Gemini) under a single Google One subscription, which keeps the cost of getting started low.

Create the folder structure

Once your tools are ready, create folders on your PC. To start, the minimum for S1 — just three folders — is enough. The top folder can be named anything; this handbook uses myproject.

myproject/
  ├── inbox/          where voice memos and rough notes land first
  ├── ontology/       the map inside your head
  │     ├── concepts/       concept files (1 theme = 1 file)
  │     ├── people/         person files (1 person = 1 file)
  │     ├── organizations/  organization files (1 org = 1 file)
  │     └── relations/      how concepts connect to each other
  └── actionlog/      a daily record of what you did

Tip: use short, lowercase, single-word folder names. You may rename folders to whatever is clear to you, but short English words make it easier to instruct the AI agent and avoid tool-compatibility issues. Keep names whose role is obvious at a glance.

This folder is your inner map (Ontology), externalized. As you progress through the framework, more folders are added — one per skill (research in S2, content in S3, finance in S5, outsourcing in S6, capital planning in S8). You do not build them all at once.

Two safety principles (before you run anything)

① Limit what the AI can see

When you let an AI agent read your folders, do not point it at your whole C: drive — it could accidentally alter system files. Point it only at myproject. Move any files you need analyzed into myproject first. “Only show the AI the folder you are working in” prevents unintended changes and deletions.

② Kick off the AI manually

Do not run the AI as a scheduled task that “wakes up in the morning and records things on its own.” Automatic tasks consume resources and cause unintended mass file generation and errors. Instruct the agent yourself, each time you work. This is the safest approach and keeps the quality of what it produces high. Scheduled and always-on background runs are not used at the S1 stage.

Step 2: input your thinking (the Data layer)

Every day or week, record your thinking into inbox.

How (example):

Record a voice memo on your phone (just talk through what you thought and noticed)
Convert it to text with a Whisper-based speech-to-text tool
Auto-save it to your PC’s inbox folder via Google Drive

Tip: manage data as MD, CSV, or JSON. PDFs and spreadsheets are costly for an AI agent to read and easily lose their structure. Keep inbox files in Markdown / CSV / JSON where possible; if you must accept a PDF or image, turn it into text first.

Tip: keep test data out of the real Ontology. Jokes, placeholders, and throwaway text go in inbox/ or a temp folder. ontology/ is where the AI reads to justify its judgments — put only real people, organizations, and concepts there. Separating the playground from production protects the AI’s judgment quality.

The point is not to write beautifully. It is to get what is inside you out, without self-censoring. One recording should cap at about 10 minutes — longer than that raises the load on the AI and lowers its organizing accuracy.

Tip (Windows): press Windows key + H to dictate at the cursor. When instructing an AI agent on something complex, put the cursor in the text field and speak your instruction instead of typing it.

Step 3: create concept files (the Logic layer)

From your inputs, create concept files in ontology/concepts/. One file = one theme:

concepts/
  ├── my-strengths.md
  ├── how-i-choose-markets.md
  ├── my-decision-criteria.md
  └── what-im-bad-at.md

What goes in a concept file:

how you currently think about this theme
your motivation and level of understanding (your state)
related events and decisions

Use the AI agent to write and extend these files. Writing them by hand is heavy — a Markdown file needs headings, tags, related concepts, people, organizations, an update history. If you format all that by hand every time, recording itself becomes a burden and you stop.

Tip: instruct from the core information. A vague instruction (“something roughly on this theme”) makes the AI produce generic content. Put the core — what to make, how you think about it, who is involved — into the very first instruction. Add details later.

The crucial part: read every file the AI makes, in full. Wherever a word is unclear, a phrasing does not sit right, or something differs from your own sense, do not leave it — have the agent rewrite it into words you understand. An unread MD, an MD you do not understand, an MD not in your own words — inside an Ontology, these are garbage. Do not let them accumulate.

AI-written documents are assembled plausibly from existing MDs and general knowledge, so a document made by the AI alone usually looks polished but has no real force. What gives an Ontology value is what you genuinely think, are unsure about, and have decided right now. Use the agent as a clean-up and structuring assistant, and keep adding your own thinking.

Step 4: record your actions (the Action layer)

In actionlog, record what you did, day by day. This is not only a place to note “did I do what the AI suggested.” Record the time you actually spent — thinking, making materials, meetings, research, email, admin, travel, publishing.

Make one file per month. Within it, one day is one block:

---
month: 2026-06
updated: 2026-06-05
---

# Action Log — June 2026

## 2026-06-05 (Fri)
> Worked: 5h | Main projects: SOVREN Framework, School

| Time | Type | Project | What | object_id | Result / notes |
|------|------|---------|------|-----------|----------------|
| 09:00-10:00 | Thinking | SOVREN Framework | Settled the open-source publishing direction | `SOVREN_Framework` | Decided to spread it as a public framework, not school-proprietary tech |
| 10:00-12:00 | Drafting | SOVREN Framework | Drafted the Manifesto page | `SOVREN_Framework` | Wrote the top message of the canonical site |

What matters is not just listing “things done.” It is that when the AI agent later reads it, which action connects to which person, organization, concept, and decision is legible. Did it, could not do it, did something else instead — record all of them. “Why it could not be done” is data too. With a daily action log, the AI can analyze your thinking patterns, use of time, procrastination habits, and which actions tend to produce results.

Step 5: record how concepts connect

Do not map the connections by hand — ask the AI agent to organize them. As concept, person, organization, and action-log files accumulate, relationships you had not noticed appear: your “strengths” influencing “how you choose markets,” “what you’re bad at” connecting to “work to outsource.”

Have the agent read everything and organize it into ontology/relations/links.json:

Review all of myproject and organize the connections between people, organizations, concepts, and the action log. Look especially for:
- concepts that relate to the same theme
- concepts tied to a specific person or organization
- continuing interests or decision patterns visible in the action log
- relationships worth adding to links.json
- existing relationships that have gone stale

Do this once a week or once a month. And always check what the agent proposes — a relationship that does not sit right, that you do not understand, or that differs from your current sense, do not leave; fix it. The relationship map is not something the AI makes for you; it is something you grow with the agent.

Step 6: analyze the whole Ontology

Once inputs, concept files, action logs, and a relationship map have accumulated, have the agent analyze the whole thing. This is not merely “asking the AI for advice.” It is re-reading the MD files you have left, your daily actions, and the connections among people, organizations, and concepts — and rebuilding your current judgment axes.

Read all of myproject and analyze the whole Ontology from the S1 point of view.
Show me:
- themes I keep returning to
- gaps between the action log and the concept files
- people, organizations, concepts that appear often
- judgment criteria I have not yet put into words
- MDs that are stale or should be merged
- the next concept file I should create
- the actions to take over the next week

Read the result, always. Turn anything unclear into words you understand. The goal is not to have the AI decide your answer — it is to update your own judgment axes, together with the agent, using the material you have left.

The cycle in one picture

thoughts, remarks, noticings
     ↓
voice memo → text (inbox)
     ↓
organize into concept files (ontology/concepts)
     ↓
record actions (actionlog)
     ↓
record connections (ontology/relations)
     ↓
analyze the whole Ontology
     ↓
decide the next action
     ↓
act → record the result → back to input
     ↓
(loop)

Tools used

Tool	Use	Cost
Cursor or Antigravity (IDE)	edit and view files	free plan available
Claude Code or Gemini (AI agent)	AI analysis, Ontology updates	a few dollars/month
Google Drive	sync files between phone and PC	free
A Whisper-based speech-to-text app	voice → text	free

Are you done?

A state in which your main axes of thinking and judgment criteria are in words, and are being updated continually.

Created a myproject folder on your own PC
Have five or more concept files
Connections between concepts recorded in links.json
Have a record of using AI to analyze your own thinking patterns
Can put into words the grounds for your daily decisions

Practice: draft a brief, then build something

As the finish of S1, build one thing with the agent — a website, a description of your company or service, a slide deck, a small app. Anything.

What we want you to experience: before you start building, construct a detailed brief in Markdown. If you tell the AI “make me an X” straight away, you get something plausible but not what you pictured. At this stage, pour everything in your head into the brief.

Four steps:

Make the brief (MD). Give the agent your requirements and have it draft the whole brief.
Read it and fix every place that differs from your image. Kill every bit of discomfort at the brief stage — fixing it in the finished product costs many times more.
Build from the brief. “Read this brief and build exactly what it says. Do not add anything not in the brief.”
Check the output, fix what is missing, and write that back into the brief. Record what you changed and why. The brief turns from a one-off spec into a file where your production judgment accumulates — the same record → analyze → update → write-back structure as the Ontology cycle.

Next: S2 — Make information usable together (once you are defined, you can decide where to fight). Currently published in Japanese: S2
Phase 2 of this axis: S6 — Manage staff with AI. Currently in Japanese: S6

This is the English edition of S1. To go deeper, the full framework — S2 through S9 and three milestones — is published in Japanese at sovren-framework.pages.dev. We run live classes on this method every week in Yokohama and publish a report after each session.

Where it is practiced

SOVREN Framework is a free, open canonical text. It is tested in live classes in Yokohama, Japan, where people implement it in their own businesses with an instructor.

About the classes (Ontology incubation) Read the session reports