How to stop re-explaining your business to every AI

Stop keeping your business context inside each AI tool. Write it down once, in plain markdown files in a folder you own, and point every assistant at that folder. The context stays the same when you switch tools, and you decide what each one can see.

Why you keep starting from zero

Chat tools remember inside themselves. What one assistant learns about your business stays in that assistant, in that account, sometimes in that one project. Open a different tool, or start a new chat, and you’re back to pasting the same three paragraphs about who you serve, what you sell and how you like things written.

That memory also lives where you can’t read it. You can’t open it, fix a wrong fact in it or take it with you. A file, you can.

One folder, three kinds of file

1. A context file at the root. One page that every assistant reads first. Keep it short and current:

# About this business

Who I serve: independent B2B consultants (replace with yours)
What I sell: a monthly retainer and a fixed-price audit
Prices: see money/pricing.md
How I work: proposals within 48 hours, invoices net 30
Voice: plain and direct, no jargon
This month: close two retainers, ship the new audit page

# Rules for any AI working here

- Read the files you need. Ask before creating new folders.
- Never send, publish or delete anything without asking me.
- Client files are confidential. Don't quote them outside this folder.

2. Facts in their own files. Prices, offers, clients and products each get a file, linked from the context file. The context file stays one page long, and each detail stays current in the one place it lives.

3. The work itself. Client notes, meetings, invoices and plans. The agent reads these when a task needs them.

Point every tool at the same folder

• Codex, Cursor, GitHub Copilot and other coding agents: name the context file AGENTS.md. It’s an open format that a long list of coding agents read, most of them on their own (Gemini CLI and Aider need a one-line setting). Open the agent in the folder and ask.
• Claude Code reads AGENTS.md too, when the folder has no CLAUDE.md. If you keep a CLAUDE.md, or run an older version, make its first line @AGENTS.md and it reads the same file. One source, every tool.
• Chat apps (ChatGPT, Claude, Gemini): add the context file to the app’s project or knowledge area. That copy doesn’t update itself, so edit the file in the folder and upload it again when it changes.
• Your phone or a server: an agent that runs in a chat app you already use, or on a machine you control, can work on the same synced folder.

Give each assistant only what its job needs

Pointing an agent at a folder shouldn’t hand it the whole folder. Decide per job. The agent drafting invoices needs clients and money. The one planning your week needs your plans. Neither needs your journal.

The simple version is folders: open the agent in the subfolder its job needs. The sturdier version is tags, so a file can live wherever it makes sense to you and still be inside or outside an agent’s reach.

Keep it current

Context goes stale quietly. A price changes, a client leaves, the month’s priority moves. Give the context file ten minutes in your weekly review, and write decisions into the folder as you make them.

Where Oknola fits

This is the job Oknola does. It keeps the folder typed, tagged and filed while you work, so any agent can read it on the first pass, and it hands each agent only the tags you allow. There’s no connector and no plugin: if a tool can read a folder, it can work on your business. What is available today is on the roadmap.

Questions people ask

Doesn’t ChatGPT’s memory already do this?

It helps inside ChatGPT. It doesn’t carry over to Claude, Cursor or whatever you use next year, and you can’t open it and correct it like a file. A folder works with all of them, and it stays yours.

How long should the context file be?

A page. Link to the details instead of pasting them in, so the file stays readable and current.

AGENTS.md or CLAUDE.md?

Write AGENTS.md, the one both can read. Current Claude Code reads it when there’s no CLAUDE.md. If you keep a CLAUDE.md for Claude-only notes, start it with the line @AGENTS.md so there’s still one source.

What about private information?

Keep it out of the shared context file and out of the agent’s scope. Give an agent the files its task needs, not the whole folder.