Cross-Brain Retrieval

When an agent connected to BlackOps answers a question about one of your brains, it can also read your others. Ask your QA brain something and the answer may quietly draw a name, a date, or a decision from your work log — because that's where the fact actually lives.

This is deliberate, and it is most of the value of keeping brains in one place rather than in separate documents. But it is worth stating plainly, because it is the kind of behavior you should hear about from us rather than notice on your own.

What actually happens

An agent asking "what do I know?" calls list_brains. That returns every brain in your account, across all your sites — name, description, aliases, and note count for each. From there the agent decides which ones to open with get_brain.

Two things follow from that:

• The agent chooses. Nothing routes a question to a fixed brain. The model reads the list and decides what looks relevant, which means cross-brain retrieval is not deterministic — the same question can pull from different brains on different days, and a brain that helped last time may not be consulted this time.
• The scope is your account. list_brains only ever returns brains belonging to sites you can access. Another BlackOps customer's brains are never in that list, and never reachable.

Why it's on by default

Closing this by default would turn BlackOps into disconnected storage. The point of a brain is that it is connected to the rest of what you know; a brain that can only ever answer from its own notes is a folder.

So the default stays open, and the control below is opt-in.

Making a brain private

Go to Brain in the admin, find the brain, and click Make private. The card shows a Private badge once it's set.

A private brain:

• is not returned by list_brains — an agent browsing your brains will not see it and cannot stumble into it
• is still readable when you name it — "check my UDX brain" still works, because get_brain resolves it by name or ID
• still appears in your own admin, so you can always find it and turn it back

Click Make discoverable to reverse it. Nothing about the brain's contents changes either way, and no recompile is triggered.

What private is, and is not

Private is a discovery preference, not a permission boundary.

It controls whether an agent finds a brain by browsing. It does not add an access-control layer, and it is not a substitute for one: anyone who can already reach your account can still read a private brain by naming it. What it prevents is the incidental case — an agent sweeping your brains for context on an unrelated question and pulling from one you'd rather it left alone.

That distinction matters most when confidential material sits next to personal material in the same account. Client work is the common case: a consulting brain and a wine brain in one account are equally reachable by default, and only one of them is something you'd want quoted back in an unrelated answer.

When to use it

Reach for private when a brain holds something you would not want surfacing in an answer to a question that wasn't about it:

• client or engagement work under confidentiality
• anything with third-party names, contracts, or internal detail
• material you keep for reference but never want quoted

Leave everything else discoverable. Cross-brain retrieval is what makes the answers good, and each brain you hide is one the agent can no longer draw on.