Assistant : One assistant beside every page - that proposes, and obeys.
The assistant sits beside every studio and designer, and in any chat you switch it on in. It works as you, with the platform's tools, and whatever would change something is shown to you first.
What it does
Assistant, point by point.
Beside the work, not in a tab
Every studio and designer has the same side bar: drop files into it, queue up what you want next, open a preview of what it made.
About this page
Switch it on and the assistant reads the page you are on. It can fill a form or press through a screen for you - after you press Do it.
It works as you
The assistant finds what you could find and opens what you could open. It is not a way around a permission.
Changes are proposals
A new list, a logged hour, a message to send: you see it as a card, and decide.
Your choice of model
Anthropic's models, an OpenAI-compatible endpoint you host, or the Claude Code, Codex and Gemini engines on your own runner - with an API key or a vendor account.
It can search the web
Through a search service you host, so questions do not carry your people's identity to a search engine.
Every token accounted for
A usage ledger says which person, agent or app spent what.
Getting started
Three steps in.
- Press the assistant button on any page.
- Ask. Drop a file in if the question is about it.
- Read what it proposes; press Do it on what you agree with.
Questions
What people ask about Assistant.
Where do our questions go?
To the model you configured, and nowhere else. That can be a model you host yourself, in which case nothing leaves your network.
What is the difference between the assistant and an agent?
The assistant is yours for the moment, on the page you are on. An agent has a standing job, a supervisor and a record of its own.