It performs the action
"Book me a desk next to Sarah on Thursday" ends with a booking, not with instructions for making one.
Product
Most workplace chatbots answer with a link to the page you were already on. This one does the thing.
In short
Jaimes is the AI assistant inside Door James. It answers questions by text or voice and carries out the action itself: find a free desk, book it, cancel it, check in, find a colleague, switch location. It searches your own knowledge base for questions about your rules, can operate the interface on request after explicit permission, and hands over to human support when it cannot help. Off by default, enabled per organization.
"Book me a desk next to Sarah on Thursday" ends with a booking, not with instructions for making one.
The same assistant works spoken – useful on the way in, when typing is the awkward part.
It queries live availability across locations, areas and resource types rather than guessing from a stale list.
Questions like "how far ahead may I book?" are answered from your knowledge base and help articles, not from a generic model opinion.
For anything it cannot do through an action, it can operate the interface itself – reading the screen, clicking, filling fields. Only after you grant permission, and you can stop it at any point.
When it reaches the end of what it knows, it passes the conversation to support with the context attached, instead of looping.
The Admin Panel shows AI usage by organization, location, user, feature and model – so the cost of the assistant is a number, not a surprise.
The same assistant is available inside the Admin Panel and helps with the configuration itself.
The standard corporate chatbot answers a question about booking a desk with a description of how to book a desk. That is worse than the floor plan the person was already looking at, because it costs a round trip to arrive at the same place.
The difference here is that the assistant has actions. It can list offices and areas, query live availability, open the booking, create it, cancel it, check in, check out, find a colleague, switch location and navigate the app. A sentence like "is there anything free near the team on Thursday afternoon" ends with a resource reserved, not with a hyperlink.
Some tasks have no dedicated action behind them – a setting in a corner of the Admin Panel, a flow nobody anticipated. For those the assistant can take the interface: read the visible screen, click, fill in fields, scroll.
That is a powerful capability and we treat it as one. It does not start on its own. The assistant asks for permission, the person grants it for that session, and it can be stopped at any point. A feature that quietly clicks through an admin interface on someone's behalf would be a bad idea no matter how well it worked.
It is not a replacement for the interface, and it is not an oracle about your organization. It answers from your indexed documents and from live data in your tenant; where it has neither, it says so and hands over to a person. It is also off until someone decides otherwise – which is the setting we would want if the question were about our own workplace.
Text or voice, in the app or in the Admin Panel. No command syntax to learn.
The assistant states the booking it is about to make, with location, resource and time.
You confirm, and the booking exists. Cancellation and check-in work the same way.
For employees
The shortest path from intention to booking, especially for the people who never open a floor plan.
For support and office management
Fewer repeat questions about rules and lead times, because the assistant answers them from your own documents.
For IT and data protection
Off by default, switched on deliberately per organization, with usage visible per feature and model.
No. It is off by default and is enabled per organization, separately for the member app and the Admin Panel. That is deliberate: an AI feature that appears without anyone deciding to have it is a problem for exactly the organizations that care most about the answer.
Control mode lets the assistant operate the interface directly – read the visible screen, click elements, fill fields – for tasks it has no dedicated action for. It only starts after you explicitly grant permission for that session, it works on what is on screen rather than in the background, and you can stop it at any moment. We treat it as a capability that has to be asked for, not a default.
From your own content: help articles and documents that are indexed for the assistant to search. A question about your lead times is answered from your configuration and your documentation, not from a general model's assumptions about how desk booking usually works.
It hands the conversation to human support, with the context of what was already discussed. We would rather it stop and pass on than produce a confident answer about your booking rules that happens to be wrong.
Yes. The Admin Panel reports AI usage broken down by organization, location, user, feature and model. Before rolling the assistant out widely, that is the number to watch.
Door James uses the assistant to answer questions and perform actions within your own tenant. Which model is used, where it runs and what the processing agreement covers is documented as part of the data protection documentation – the appropriate place to check it before switching the feature on.
Thirty minutes with someone who works on the product. Bring your floors, your booking rules and the awkward case you think will break it.