Set it up so nothing
you do is ever lost.
The whole setup, start to finish. Watch the walkthrough, or follow the written build below.
Good work I do with AI used to vanish the second the chat ended, and a backup only saves what I wrote down. So I built the seven-part setup that fixes both: a plain-text vault you own, an AI rulebook that captures your decisions as you work, automatic GitHub backups that keep every version, and a private search engine over everything you have written, synced to every machine. Build it in order, and you do not need all seven on day one.
Watch the build top to bottom, or read it below. Same seven pieces either way.
You do good work with AI. You make decisions, solve problems, have ideas. Then the chat ends and it is gone. This is the setup that makes sure that never happens to you again.
The vault, where your work lives
- 01Download Obsidian at obsidian.md. It is free, on Windows, Mac, and phone.
- 02Open it and create a vault. A vault is just a folder you pick. Your notes live there as plain text files that you own. Nothing is locked in someone else's cloud.
- 03That folder is the foundation. Everything else points at it.
That is the whole start. You now have a place to keep your work that you control.
Claude Code and the rulebook
This is what turns a notes folder into a system.
- 01Install Claude Code at claude.com/claude-code.
- 02Point it at your vault folder (open that folder in your terminal and run
claude). - 03In the folder, create a file named CLAUDE.md. This is the rulebook your AI reads before it does anything, every time.
# How to work with me
## At the start of every session
- Read today's daily note. If it does not exist, create it.
- Tell me briefly where we left off.
## While we work
- When we make a decision, write it down in the right note.
- When we solve a problem, capture how.
## At the end
- Write a short log of what we did and what is next.Now capturing your work is the default. The AI does it because the rulebook tells it to, not because you remembered.
The commands
These are the deliberate captures you trigger. In Claude Code you set them up as simple reusable prompts (slash commands). Three carry the whole load.
At the end of a good conversation, the AI reads the whole thing, pulls the decisions, learnings, and open tasks, and files them as a clean note. You do not sort it. It sorts itself.
Saves the work, commits and pushes the code, and syncs what you learned into memory. You do not close a laptop on unsaved work.
When a chat gets long, the AI writes a short brief: where you are, what is next. You paste it into a new chat and keep going without losing the thread.
Read back over this whole conversation. Pull out the decisions we made,
the things we figured out, and any next steps. Write them as a dated note
in my vault, filed by topic, and link it from today's daily note.GitHub, a backup that remembers every version
- 01Make your vault a git repository and push it to a private GitHub repo. Git is a backup with a memory: it keeps every version, so you can rewind to any point in time.
- 02Optional but worth it: a small scheduled job that commits and pushes every 15 minutes, so you never think about it.
cd /path/to/your/vault
git add -A
git commit -m "auto: $(date)" || exit 0
git pull --no-edit
git pushSchedule it every 15 minutes (Task Scheduler on Windows, cron on Mac). Keep the repo private. Never commit secrets (see the security note at the end).
Docker and LightRAG, ask your whole history a question
This is the part people think is hard. It is not.
A notes app finds the exact word you type. This finds the meaning across all your notes and answers in plain English, with the source. It is your own private search engine over everything you have written. The everyday command for it is /lightrag-query: before you decide anything twice, you ask what you already decided.
- 01Install Docker Desktop.
- 02Run LightRAG (open-source) in a container, bound to your own machine only (localhost). One command and it is up, and it never leaves your computer.
- 03Give it a model provider for the answers and the search math (for example an OpenAI key for embeddings and answers, and optionally a reranker). Put your keys in its config file, not in your notes.
- 04Point it at your vault folder so it reads your notes, and let it build its map.
- 05Ask it questions. It finds the closest ideas by meaning, ranks them, and writes the answer with the note it came from.
Keep it on localhost. Your whole history becomes searchable and it stays private.
Automation, so it runs without you
Set two scheduled jobs and forget them:
- 01Every 15 minutes: commit and push your vault to GitHub (Part 4).
- 02Every few hours: feed any new or changed notes into LightRAG and re-index, so your search engine is always current.
Now the system captures, backs up, and re-indexes on its own, whether you are at the desk or not.
Cross-device, the same brain on every machine
Put your vault folder in a synced location (OneDrive, Dropbox, or iCloud Drive) and rely on git as the second layer. One layer copies your files between machines almost instantly. Git keeps the versions straight on top of it. Close your laptop, open your desktop, and it is the same brain: same notes, same memory, same history.
Layer on layer
A rulebook that writes as you work. A vault you own. Commands that file everything. GitHub backing it up every fifteen minutes with full history. A private search engine over your entire past. Automation running it hands-off. On every machine you touch. It is not one save button. It is a failsafe, layer on layer, so no work you do ever gets lost.
Before you ask
Want this in your business, not just for yourself?
The build above is the personal version. Putting it inside a business, so it holds what is scattered across your whole team and fits the tools you already use, is a different job. That is the part I do with you.
Set it up once and nothing you do is ever lost. That starts with a free call, no pitch.