The message box has two menus, and they answer two different questions. + sets where and how a chat works: which data it uses, which workspace or connection it is in, and whether it checks its own work. / says what to do: make a dashboard, write a report, lay out a plan. You can ignore both and just ask a question; they are there for when you want more control.
What you’ll need
- A chat. Some rows only appear in an open chat (Quick question, Recap, Save this chat as an agent), and a row your plan doesn’t include is listed but won’t run.
Open a menu
- Click + at the left of the message box. On a phone it opens as a sheet from the bottom, with a search box (“Search data, workspaces, connections…”) at the top.
- Or type / as the first character in the box. The list narrows as you type, so
/repfinds Report. Press Enter to run the highlighted row.
Where a group is long (many workspaces, agents or connections), it shows the first four until you type something to search for.
What + holds
- Upload a file — Excel, CSV or Parquet; Crunch can read it once it lands. See Upload files.
- Choose datasets — your uploads, linked tables and your organization’s data. See Datasets.
- Workspaces, Your agents and Connections — pick one and the chat uses it from your next message. See Workspaces, Agents and Connections.
- How it works — three switches:
- Reflect checks its own work before answering. It is slower.
- Delegate hands a job, such as testing why a number moved, to an agent of its own. It is slower too.
- Web search lets Crunch search the web for this chat. It is in the Pro+ plan, and it isn’t listed when your organization has turned web search off.
- New connection and New workspace — start one on its own page.
A switch that is on also shows as a chip beside the +, with a cross to turn it off.
Heads up: Reflect and Delegate are settled by a chat’s first message. After that they show greyed out with “Chosen with a chat’s first message; start a new chat to change it”. Web search is different: you can change it for each message.
What / holds
- Dashboard — a real dashboard, opened beside the chat; each message adds or changes its tiles.
- Report — a real doc, opened beside the chat; each message writes or rewrites it.
- Plan — works out the steps as a doc beside the chat, and does none of them until you say. Say what to plan after it:
/plan why did revenue drop in March. - Notes — writes what the one attached dataset means: what a row is, each column, terms and caveats. Attach exactly one dataset first.
- Quick question and Recap — see below.
- Your own agents, to run one in this chat.
- New agent — a draft of your own; the co-pilot writes its instructions with you. Type a name after it:
/new-agent weekly-recap. - Save this chat as an agent — a draft that uses what this chat uses, started from what you asked here.
The first three are explained in Dashboards and docs with their own chat.
Typed forms
Any row can be reached by typing its name after the slash. To attach something by name:
/workspace-<name>,/agent-<name>or/connection-<name>. Names may contain spaces:/workspace-Acme Corp.
To run a command that takes a word: /new-agent <name>, /save-as-agent <name>, /plan <what to plan>. If you pick one of these from the menu without the word, it waits in the box for you to finish it.
Quick question and Recap
Quick question asks about the conversation on the side. It opens a panel on the right of the chat (a sheet on a phone), and neither your question nor its answer is added to the conversation. In the panel, Send asks it, Insert puts the latest answer into the message box (unsent) so you can build on it, and Clear closes that question. The panel’s questions are saved with the chat; with the panel hidden, a Quick questions link under the chat shows how many you have.
Recap gives a few short lines on what was asked, what was found, what was made and what is still open. Like a quick question, it is asked on the side and not added to the conversation. Once a chat has two answers, a Recap link also appears under it.
How to tell it worked
A command either opens something (a dashboard or doc beside the chat, the quick-question panel) or switches something on, which shows as a chip. A row that can’t run says why instead of doing nothing.
Related
- Dashboards and docs with their own chat — what /dashboard, /report and /plan open.
- Agents — save a job and run it again.
- Workspaces — one area of work with its data and notes.
- Models and effort — the other chips beside the box.