Features · The helper
A helper already in the room.
You do not bring an AI and you do not learn prompts. The helper is part of the plan, and it uses the same tools an outside AI would use.
It looks first, then does.
The helper works inside your workspace as its own assistant, with the same permissions, plan limits and history as any AI you connect. It looks before it changes things, says in a sentence or two what it did, and says so when something came back empty instead of making it up.
- Reads your workspace before it edits it
- Short answers in the language you wrote in
- Never invents data: an empty result is reported as empty
Its toolbox
What it can use
Your tools
Notes, tasks, lists, reminders, calendar, contacts, documents, invoices, forms, files and more, the same tools as in the app.
Your apps
The apps you installed, through the same door.
Its notes
A few facts you told it or it noticed, kept on a page you control.
Quick sizing
Small model, right size.
Routine jobs
- A fast, small model (Haiku 5.5)
- Notes, tasks, calendar, lists
- Cheapest per message
When it counts
- A bigger writing model (Sonnet 5.5) for long writing
- The biggest (Opus 5.5) only when you ask for the best
- You never pick; asking for "the best model" does
Honest labels
What to expect.
- LiveLimits on every message
At most 6 tool steps per message, a cap on how much one message may write and spend, and a daily ceiling for each workspace.
- Pre-betaTest mode
A practice workspace can run in test mode: the helper then gives clearly marked practice replies, says "Test mode" in each one, and every check still runs.
- PlannedA smaller model of our own
We are looking at open-weight models for simple jobs. Nothing is decided and nothing is promised.
Questions
About the helper.
Does it include an AI?
Yes. The helper is built in and is part of the plan: you do not open or pay for an AI account on the side. It runs on Claude models: a fast small one for everyday jobs, a bigger one for long writing, and the biggest only when you ask for the best.
What does the helper do?
It reads your workspace and uses its tools: notes, tasks, lists, reminders, calendar, contacts, documents, invoices, forms, files and the apps you installed. It keeps its answers short and tells you what it did. See the helper inside.
Will it make things up?
It is told never to invent data and to say so when a tool returns nothing. It can still be wrong, as any AI can, so read drafts before they go out. That is why sending always waits for your tap.
What is test mode?
While a workspace is in test mode the helper gives clearly marked practice replies instead of real AI answers, and says "Test mode" in each one. Everything around it still works, so people can try the flow. Early-access workspaces may start in it.
Which model is it?
Claude models. A small, fast Haiku-class model does the everyday jobs, a Sonnet-class model does long writing, and an Opus-class model only when you ask for the best. See the helper inside.
What do I say first?
The empty chat offers three starters that fit your kind of workspace. A good first one is "What is on my plate today?". See what to say.
What if an email tells the helper to do something?
Text that comes from mail, web pages or forms is treated as data. The helper tells you what it asked for and lets you decide.
What does the helper send to a provider?
Your message and the parts of your workspace the helper reads to answer it go to the model provider, as with any AI. Your voice goes to the speech service. The privacy statement says what is stored.