The interface here is built unusually: it is a showroom, not a control panel. You watch what is happening and give commands to the agent, and it changes the settings. That is by design — and, honestly, you get used to it faster than you expect.
Eight screens
- Home — the summary: how much material and how many events arrived, what needs attention, how the quotas are spent.
- Monitors — what you watch and how it is doing: value history, charts, source health.
- Events — the feed of changes with evidence and materiality ratings.
- Inbox — the collected material: everything that arrived from the sources, readable, with a way to open the source.
- Published — drafts and finished publications with their history.
- Delivery — where and how messages went: statuses, answers, retries.
- Analytics — the numbers: volumes, materiality, AI spending, source quality.
- Scripts — your Python scripts: what they do, which step they are wired to and how the recent runs went.
Read-only
There is not a single editing form in the interface. A human can do exactly two things: mark an event as read and add it to favorites. Everything else — from adding a source to changing a threshold — is done by your agent.
The reason is simple: configuring the watching consists of dozens of connected little decisions. When the agent makes them, there are no “half-configured” states, forgotten fields and lost edits — and you do not spend an evening clicking through settings.
Command cards
Next to every item there is a command button. It prefills a ready phrase into the chat with the agent: “check the monitor now”, “prepare a draft for this event”, “pause the monitor for a day”. All that is left for you is to send the message.
When the chat is not connected, the button copies a ready command for calling the product — it can be sent by hand, from your own system. The product’s capabilities are the same in both cases; only the way of giving the command changes.
Links to details
Every event, monitor and publication has its own address. An event opens with a link like /?view=events/42, a monitor — /monitors/{key}, an item — /inbox/123, a publication — /published/12, a delivery — /delivery/7. Such a link can be forwarded to a colleague: they will see the same card with the same evidence.
On a wide screen the address is visible in the browser bar — convenient to copy. On a phone the address bar is usually hidden, but the “back” button works: it returns you to the list you came from.
Live updates
An open screen refreshes itself: new events, material, deliveries and publications appear without reloading the page. Frequent updates are merged so the interface does not jitter. In the side menu, the section you need gets an unread badge — so you see something arrived even when you are on another screen.
Clear states
Instead of codes and service words, the screens say what is happening. A source: “Active”, “Failure streak”, “Error”, “Paused”, “Awaiting first check”. A value: “No changes”, “Changed”, “Below threshold”, “Flicker”. A rating: “Unscored”, “Noise”, “Worth knowing”, “Material”, “Existential”. A delivery: “Queued for retry”, “Suppressed”.
This is not decoration: such a name immediately tells you what to do — no need to check the documentation or guess what an error code means.
What to do when it is empty
An empty screen always explains what it is waiting for: “monitors are created and configured by your agent — tell it what to watch”. That is why the first launch does not look like a breakage: it is clear what the next step is, not that “nothing works”.
Next → Analytics, Search & Exports