Getting started
Quickstart
Ten minutes from an empty workspace to a running system — with a real example you can follow word for word.
You need an account and about ten minutes. Nothing to install.
1. Create a workspace
Sign up, and you land in your first workspace. A workspace is a sealed container: its own data, its own files, its own assistant, its own members. Most people keep one per project or per client.
Give it a name that means something to you — you’ll be switching between several soon enough.
2. Ask for something real
Open the assistant and describe what you want in ordinary sentences. Don’t design a schema. Describe the situation.
Track our inbound leads. For each one I want the company, the contact, where it came from, what stage it’s at, and when we last spoke. Show me a board grouped by stage, and flag anything we haven’t touched in two weeks.
Watch what happens. The assistant will:
- create a Lead type with the fields you described,
- create a board view grouped by stage,
- add a second view or a filter for the stale ones,
- put the board in your sidebar so you can find it again.
It narrates as it goes. If it makes a choice you don’t like — a field name, a stage list — say so and it changes it. It’s faster to correct than to specify everything up front.
3. Put some data in
Three ways, all fine:
- Tell it. “Add a lead: Northwind Trading, Dana Ruiz, came from the webinar, stage qualified, last spoke Tuesday.”
- Upload a file. Drop a spreadsheet or CSV into your workspace drive and say “import this into Leads.” It will read the file, match the columns, and tell you what it skipped.
- Type it in. Open the type from Data structures and add records by hand.
4. Open what it built
Everything the assistant made is a real object you can open:
| Where | What you’ll find |
|---|---|
| Sidebar | The views it created, in the order it placed them |
| Data structures | Your types, their fields, and every record |
| Views | Each saved view, its components, and its data sources |
| Workers | Any AI colleague it defined for you |
| Workflows | Multi-step processes it set up |
Open the board. Change a card’s stage. The underlying record changes with it.
5. Take over by hand
The assistant is a fast pair of hands, not a wall. You can rename a field, delete a view, edit a record, or rewrite a workflow step yourself at any time — the assistant picks up from wherever you left things.
Two habits worth forming early:
- Correct it in the moment. “No — stage should be a fixed list: new, qualified, proposal, won, lost.” It will migrate the field and keep your data.
- Tell it your rules. “We never contact anyone twice in one week.” Constraints like this get remembered and applied later, without you repeating them.
6. Make it run without you
The last step is what turns a dashboard into a system.
Every Friday at 8am, look at leads we haven’t touched in two weeks and send me a short brief on the ones that were warm.
That creates a worker — a named AI colleague with one job — and puts it on a schedule. It runs whether or not you have the tab open, and reports back when it’s done. Open Workers to see it, run it manually, edit its brief, or turn it off.
What to read next
- Workspaces — how to organise more than one.
- The Assistant — how to steer it, and what it remembers.
- AI workers — naming colleagues, scheduling them, and putting several in one chat.
Asking well
A few patterns that consistently produce better results:
Describe the outcome, not the implementation. “I want to know which invoices are late” beats “create a table with a due_date column and a computed overdue boolean.”
Give one example. A single real record teaches the assistant more about your data than a paragraph of description.
Say what you don’t want. “Don’t create a separate table for contacts — keep them on the lead” saves a round trip.
Work in passes. Get the data right, then the view, then the automation. Asking for all three in one sentence usually works, but reviewing each stage catches drift early.