Witbitz Β· agents

An agent is a participant, not an endpoint.

It lives in the room rather than behind a request. That difference is what the rest of this page is about β€” how it behaves, what it arrives with, and how your own functions become things it can do.

How it behaves

The difference between an API call and a colleague is everything below.

It persistsThe room is a ledger, not a session. Close the tab, come back in a week β€” the agent still knows the whole thread.
It answers on its own timeNobody has to sit and wait for a spinner. It works while you're gone and the reply is there when you're back.
It asks before it actsAn agent proposes; the runtime checks policy; an authorized human approves or edits; then it executes. Enforced by the platform, not by asking a model nicely.
It shares its workWhat it draws, renders or fetches is a widget everyone in the room sees β€” the same state, not one person's screenshot.

What it arrives with

Tools the agent already has, so a room is useful before you wire up an endpoint of your own.

πŸ“„ Read what you sharePDF, Word, Excel and CSV come in as text it can actually reason over. Images too β€” it perceives them natively.
πŸ—ΊοΈ Show a real mapA vector basemap with proper labels, place search, and routes β€” rendered as a widget, not a screenshot.
πŸ“Š Draw the chartCharts and diagrams as live SVG widgets, inline in the conversation.
πŸ“ Write the documentGenerates a shareable PDF from its own output β€” including right-to-left scripts.
✈️ Search the real worldLive flight and place search, returned as pickable cards rather than a wall of text.
🎨 Make thingsGenerates and edits images, renders a video recap, and in a call it can sing your meeting back to you.

Your functions, the same way

Declare your endpoints over HTTP or MCP and the gateway calls them β€” metered, with egress governed. There is no first-class/second-class split between the tools above and yours: they are declared identically and the agent cannot tell them apart.

The approval gate applies to both. An agent proposing pipeline.setStages against your API goes through exactly the policy check as one proposing to draw a chart β€” which is what makes it safe to give an agent a tool that changes something.