Connect

No connector. It’s a folder.

Claude, Codex, Cursor, or whatever you switch to next year: if it can read a folder, it can work on your business today. There is no plugin to install, no API key to mint, and no official integration to wait for.

The usual friction

Most tools make you wait for permission.

Your work lives behind someone’s API. To let an assistant near it you need a connector that a vendor decided to build, scopes that vendor decided to expose, and a plan tier that includes it. When you change assistants, you start that queue again.

Plain files skip the queue. Every serious agent already reads them.

Three ways in

Point it at the folder. That’s the setup.

Pick whichever one you already use. The folder does not change, so you can use all three on the same files.

On your desktop

Claude · Codex · Cursor

Open the agent in the folder and ask. It reads the files it needs, writes the ones you ask for, and everything it does lands as markdown you can read afterwards.

On your phone

A contact that knows the files

The agent answers in the chat app you already have open. It holds the tags you gave it and nothing else, so you can ask it about a client in a queue at the post office.

On a server, 24/7

OpenClaw · Hermes

Run a harness on a machine you control, working the same synced folder while you sleep. Your agent, your infrastructure, your files.

What it reads

Legible on the first pass.

A pile of notes is readable, not legible. Because Oknola keeps every file typed, tagged and described in OKF, the agent knows what a file is before it reads a word of it, and where to look for the next one.

That is the whole trick. No retrieval service, no embedding job, no sync that can fall behind: the files are the source of truth and the files are the interface.

One file, as the agent sees it
---
type: Meeting
tags: [Meeting, BusinessOS]
description: Scope call, two locations
client: [[Fluff & Buff Grooming]]
date: 2026-09-11
---
Pickup windows, how drivers are vetted,
one price. Proposal due Friday.

The boundary

Connected is not the same as unlocked.

Pointing an agent at the folder does not hand it the folder. Tags are the permission model: give it your clients and your ledger and it never sees your journal, whatever folder that journal sits in. Run one per job if you like, each with its own tags.

Agent off, your files never leave your device. Agent on, you chose the scope.

The Business agent in the demo. It answers which design partner is stalling and files the follow-up. Then, asked how you slept this week, it refuses: journal pages are tagged Daily_Log, a Life tag, and this agent is not allowed Life files. Its scope shows business and shared tags allowed and life tags dashed out, and the switches show Business and Shared on, Life off. — The same agent answers about clients and refuses the journal, because Life is switched off. Try it in the demo →

Why it keeps working

We don’t pick your AI.

Compatibility here is not a partnership we signed; it is a property of plain files. Nothing on our side has to be updated when a new model ships, and nothing breaks when you drop one assistant for another. The folder you built this year is the context you hand to whatever you run next year.

What an agent writes has to conform to the schema too, so the classic failure, an assistant cheerfully reorganizing your notes into rubble, does not land. Everything is plain text, so every change is inspectable and reversible.

Bring the agent you already trust.

The free app comes first, on macOS and Android. Early access opens in waves.