Watching your agent work

Most of the time your agent works without you. This page is about the screens you use when you want to look.

Conversations

Conversations is the inbox. It lists every customer conversation your agent has handled, across WhatsApp and web, with a running total at the top of the page.

Each row tells you six things:

• Status: whether the conversation is open, waiting on the agent, escalated, or resolved
• Customer: the customer's handle, partially masked
• Latest message: the most recent thing said, so you can scan the list without opening anything
• Lang: the language of the conversation, when it could be determined
• Channel: WhatsApp or web
• Last active: how long ago something happened in this conversation

Two of those deserve a note, because they behave differently from what you might expect.

The language cell is blank when the language could not be determined. No chip beats a confident wrong one, so nothing is guessed there.

The latest-message cell reads "No messages mirrored yet" for a conversation that exists but whose messages are not yet available to display. That is a real state and not an error.

Filtering

Three sets of filter chips sit above the list: State, Channel and Active. Active is a time window, so choosing "Last 24 hours" and "Last 7 days" together gives you seven days rather than nothing, since the widest selection wins.

Filters apply to your whole history, not just the conversations currently on screen. This matters when you have a lot of them: a filter that only looked at the visible page would quietly report "no results" for conversations that exist just past the end of it.

If a filter genuinely matches nothing, the list says the filters matched nothing rather than telling you that you have no conversations.

Opening one

Click a row and a detail panel opens beside the list with the same facts plus a message count and the latest message in full. From there, "Open full transcript" takes you to the conversation itself.

The transcript

The transcript page has two tabs.

Transcript is the conversation as it happened, in order. Every turn is labelled with who said it and when:

• Customer for the person your agent is talking to
• Agent for your agent
• A person's name for a message a human on your side sent, which is covered in Taking over a conversation
• Tool for an action the agent took rather than something it said

Human messages carry their own delivery state, shown on the message rather than in a notice somewhere else. A message that reached the customer says "Delivered". A message that did not says so plainly and tells you to send it again. That state stays on the message after you send another one and after you reload the page, which is the point: a warning that vanishes is a warning nobody reads.

If the conversation is long enough that older messages exist beyond what is shown, the page says so rather than letting you assume you have reached the beginning.

Details is the same conversation as a fact sheet: status, customer, channel, whether it is currently being handled by the agent or by a person, language, when it started, when it was last active, how many messages it holds, and any escalations it produced. Escalations here link straight to the Handoffs screen.

Two fields on this tab follow the same honesty rule as the rest of the console. Token usage appears only when it was actually measured, rather than showing zero for a conversation that plainly used some. An ended conversation shows when and why it ended; one that has not ended shows neither.

The agents roster

The Agents screen names the AI employees running in your workspace. Until you have seen it, it is easy to look at a list of conversations and infer that something produced them without ever being told what.

Each agent card shows its name, its status, a description if it has one, how many conversations it has handled, and when it was last active. An agent that has never handled a conversation reads "No activity yet" rather than borrowing its creation date and looking busier than it is.

The status shown is the provisioning state from our records. It is not a live health check, because none is performed, and the screen does not pretend otherwise.

The roster is read-only. You can see your agents; you cannot create, rename, configure or remove one from this screen, and there is no other screen that lets you. Changes to an agent are something you ask us for. This is the absence you are most likely to notice first, which is why it is stated here as well as on What is not here yet.