Step 9

AI and data

What it is

Roles say what people see. AI and data says what each AI may get. Without any setting nothing changes: any AI may get everything, as before.

AI roleCoversLeaves out
Any AIonlybrain AI (the chat, summaries, descriptions and transcripts on the onlybrain account), members' apps through the connector (Claude, ChatGPT, Claude Code), the company's own AI and the local AI.Nothing, this is the default with no limits.
Company's own AI onlyAI the company connected itself in Organization AI: its own API key (Anthropic), a runner with Claude, its own server in a cloud. And the local AI.onlybrain AI and members' apps through the connector (Claude, ChatGPT, Claude Code), even when the company pays for their subscription.
Local AI onlyModels on the company's server, connected by a runner (the company's graphics card). The data never leaves the company.No cloud at all: onlybrain AI, members' apps, the company's own API key or a runner with Claude.

The app shows the same overview at both settings (in the brain and in Organization AI → Data).

Set it up

  1. In the brain, open Settings, the AI and data part (brain managers only).
  2. Click Set what each AI may get.
  3. At the top, pick The AI role being set: Any AI or Company's own AI only. The local AI always sees the whole brain, and the data does not leave the company.
  4. In the tree of folders and files, choose Sees or Hidden for each item, as in the role editor.
  5. In the What the AI sees preview, check how many files each AI gets (the company's local AI, the company's own cloud, onlybrain's AI, the members' apps).
  6. Save. Discard changes returns to the last saved state.

What is hidden from any AI goes only to the company's own AI and the local AI. What is hidden from the company's own AI too goes only to the local AI. A subfolder can only make the protection stricter: the strictest setting of the folder and all folders above it applies.

The setting is saved in the file .onlybrain/ai.yml, and its changes are in the brain's history. The brain refuses an invalid file. Should one appear anyway, the whole brain counts as local AI only until a manager saves it again.

What it guards

The server enforces it, not an instruction to the AI. The chat, summaries, document descriptions, meetings, processes, the connector and runner jobs never send protected files to another AI, not even when asked to.

  • Without a connected local AI, files for the local AI only are not summarized or described.
  • The chat over all brains leaves out a brain the chosen AI may not read, and the answer says so.
  • In brains with protected data, the internet in the chat is off by default.

It does not limit people: a person whose role sees a file still sees it and can open it in their own tool too. So it is a good idea to hide protected folders from the roles that work with a cloud AI as well.

A minimum for the whole organization

In Organization AI → Data, the organization's admin sets the Least sensitivity of every brain. No brain then sends data to an AI below that level, whatever its manager sets. In the brain this shows as "The organization requires at least: …". The same place shows Where the runners' data goes: which runner is local AI and which goes to the cloud.

More about the organization's AI in the guide Organizations and the organization's AI.