Step 11

The chat in the app

Open the chat

  • The Ask the brain button in the bottom right of every page, or the shortcut ⌘J (Ctrl+J on Windows). The same shortcut closes the chat.
  • The address /chat opens the chat in the brain used last. Open in a new tab in the chat's header moves the conversation into a tab of its own.
  • On the Brains page, Ask all brains opens the chat over all brains right away; next to it is a button for the chat without a brain.

The organization's admin turns the chat on and off (Chat in the app in the organization's settings). When the chat is not available, the window says so.

What the chat works over

ChoiceWhat it can do
one brainsearches and reads the brain by the role, can use skills, agents, commands and save an answer into the brain
All brainssearches and reads every brain the account is a member of, each by the role there. It only reads, an answer cannot be saved into a brain. A brain the chosen AI may not read is left out, and the answer says so.
without a braina plain chat about anything, the AI sees the content of no brain

The choice is made in the chat's header and remembered in the browser. A change starts a new conversation, an open conversation keeps its own choice. Commands such as /init, /plan, /agent or /background work only in the chat of one brain.

Steps of the work

While answering, the chat shows what it is doing: "Reading clients/acme.md…", "Loading the skill…", "Thinking…", "Writing the answer…". Below are the steps with their times: what the AI searched for, read and found, and short notes on why it goes on. When done, the steps fold into one line (say "5 steps · 12 s"), a click unfolds them. Esc stops the answer.

Context gauge

Next to the input field is Context N %: how much of the model's window the conversation takes. A click shows the breakdown: system (rules and the brain's card), tools, conversation history, files and tool results, and free space. From a certain size, the older part of the conversation is compacted by itself. Close to the limit the gauge offers Compact (the same as the /compact command).

Which AI answers

A button in the input field shows how thoroughly the AI should work and where it answers. More thorough work uses the limit faster, and the highest levels come with the Pro plan and higher. The choice applies from the next message.

Where the organization routes the chat to its own AI, it can also be chosen where to answer: Local AI (the company's graphics card), onlybrain, or Organization AI (the company's own key). The local AI is offered right after a brain opens. Each answer says which AI answered, for example "Local AI · model", and when the fallback answered instead, it says that too. Data labelled in AI and data for the company's own AI only or the local AI only never goes to another AI.

The chat on the onlybrain account uses a daily or monthly limit that the whole organization shares. The chat shows how much is used (Chat usage).

Internet in the chat

Where the organization has the internet in the chat on, the chat can search the web and read web pages. The answer then lists Sources from the web with links. Only the search query and the page address go to the web, the content of the brain does not. In brains with data for the company's own AI only or the local AI only, the internet is off by default. The organization's admin sets it, see Organizations and the organization's AI.

Typing and commands

  • / offers the chat's commands and the role's skills, @ files, folders and agents.
  • Enter sends, Shift+Enter adds a new line, ↑ brings back earlier questions. Shift+Tab turns on plan mode: the AI first proposes a plan and carries it out only after Carry out the plan.
  • Save to the brain saves the answer as a proposed note in the inbox/ folder.
  • Background (or /background) hands longer work to the AI in the background. Its card shows whether the job waits, runs or is done, and it can be cancelled.

How to ask good questions is in the guide How to ask the brain.