Your Dashboard
The first screen in the app, and it is about your account rather than any one agent. Five blocks, in this order.
Your agents, as a row of cards you can scroll.
Each card carries the kind, the address, whether it's live, and a setup bar showing how far through the six activation stages it is. That bar is the one place two agents can be compared at a glance — see an agent's Overview for what the six stages are and what each is worth.
New agent is the first card, so it is in the same place whatever your account holds. The cards do not wrap onto more rows as you add agents — they scroll sideways in their own track, so this block stays one card tall and Needs you, the only block here that is a list of things to do, stays on the first screen. On a wider screen an arrow appears on each side there is more to reach.
An agent somebody shared with you sits under Shared with you, marked Read-only, with no figures — it's read through another account's grant, so none of your numbers include it. That list wraps onto as many rows as it needs rather than scrolling, because it is short by construction and nothing about it changes as your own agents multiply.
This week. Conversations (and whether that's up or down on last week), questions counted, how many raised nothing to teach, and new people. Every one of these adds up every agent you own and says so underneath.
Plan and usage. Your plan, your trial countdown if you're on one, and a meter per agent. The meters are per agent because the limits are: your plan's source and word ceilings are enforced against one agent's material, so a single combined bar would be a picture of a limit that doesn't exist.
Needs you. One queue over every agent: questions waiting to be taught, agents waiting on their identity check, agents that were turned down, and agents that are still drafts. Every row names its agent and opens that agent's own page, where the thing to do actually is.
Quick links. Add knowledge, train, install snippet, and — for a Support or Team Agent — invite a teammate.
What is not here any more
One agent's journey, publish state and teach box. Those moved to an agent's Overview in September 2026. This page had been both things at once: fleet-wide counts standing beside one agent's checklist and its teach queue, with a caption under the queue explaining that it covered one agent while everything around it covered several. The teach box in particular had to move — answering there writes a correction against one agent, so a queue pooled across agents could teach the wrong one.
Your recent conversations. Activity is where a conversation is read now — every one as a run, with what was asked, which agent took it and how it ended, which is more than the four rows this page used to show. It's one press from the Overview ⌄ menu on this page's title. The Most cited tally stayed where it was, inside This week.
Trials, and why this page says so
Making your first agent starts a 14-day Founding trial. Until this page, nothing in the product mentioned it, so people met it by having it end. Plan and usage now says how many days are left, every day of it. When it ends nothing is deleted — your limits go back to Free.
Two columns most of the time
The narrow column on the far left is the spine: Dashboard, Agents and Knowledge, with search, invite, settings and your account below them. It shows icons until you point at it — or Tab into it — then opens to show the words. Pin it and it stays open.
Dashboard and Knowledge are the spine and the page, nothing else. Their other pages live on the title: it reads Overview ⌄, and opening it lists Overview, Insights and Activity. The Knowledge Hub's lists Sources and Add — two, because the Hub is the whole account's library and has no one agent in context. It works from the keyboard like any other menu.
Agents is the one section with a second column, because it's the one with enough pages to need it: Build, Train, Test, Publish, with the agent switcher above them.
Pressing Agents opens a list of every agent on your account — one card each, with its kind, its address, and whether it's live or a draft. Open one and it becomes the agent the second column is about; the switcher above that column is the shortcut to the same place. New agent is on the list too, and opens the same guided setup you walked the first time.
An agent you were invited to shows under Shared with you, marked Read-only, with no way to make a new one — agents belong to the account that created them.
Build is a page as well as a stage. It gives you one card per surface your agent is assembled from — Profile, Knowledge, Voice & rules, Lists, Actions and, for a Support or Team Agent, Escalation — each showing what it holds right now: links on your profile, sources in the index and their kinds, topics scoped, lists and actions set up, and when your agent hands a conversation to a person. It's a glance, not an editor: Open goes to the real page, ? to that page's guide. A personal twin sees five cards rather than six, having nobody behind it to hand a conversation to.
One figure isn't there yet, and the card says so: which of your topics your answers actually lean on. We know which sources an answer cited, but matching those back to your scoped topics isn't a measurement we can make honestly today.
The preview is a private chat with your agent, on the right, beside whatever you're editing. It's there while an agent is open — not on Dashboard or Knowledge, which are about your account rather than one agent, and not in the Sandbox, which is already a chat with that same agent. It starts closed; open it from Preview in the top bar, and pin it to keep it. Its tabs show the agent in each placement you have — the chat widget with the launcher and card a visitor presses, the chat bar as its two cards. Each tab keeps its own conversation, so you can move between them and come back to what you were asking, half-typed question included; Reset in the ⋯ menu is what clears them. Nothing you ask it counts toward your message limit, reaches your Insights, or lands in your teach queue.
Publish is in the top bar only while an agent is open. It publishes one agent, and on Dashboard and the Knowledge Hub there is nothing on screen saying which — the same reason the preview isn't offered there either. It's away on the agents list for a different reason: that page is about all of your agents, so there's no one agent for it to publish. Open an agent and it's back.
The agents list says which agent the preview is talking to, in a line above the cards, wherever the preview is a column beside it: the list is about all of them and the chat beside it is about one, so it names it rather than leaving you to guess from a highlighted card. On a phone the preview opens over the page rather than beside it, so there's nothing beside the list to name and the line isn't there.
On a phone all of it folds into one drawer behind the menu button, in column order: the spine and your account first, then — under Dashboard, Overview, Insights and Activity as their own rows, so you don't have to close the drawer to find the menu on the page title — and under Agents, that agent's pages with the switcher above them. Your account is also one tap away in the top bar. Escape closes the drawer and the preview sheet and returns you to the button you opened it from. While either is open the page behind it stays where you left it rather than scrolling under your finger, and the drawer or sheet scrolls on its own. If you have opened something over one of them — your account menu, or a confirmation — Escape closes just that, and the drawer stays where it was.
Every part of the app has exactly one home in this list, always one click away at the address it has always had.
Build, then Train, then Test, then Publish — the order the work happens in, the same on every visit. Learn it once instead of re-reading it.
Under Agents, the others show their name until you click one, and stay open once you do — on a phone the rails are a full-height drawer, and four open stages turn navigating into scrolling. The stage you're in has no arrow, because it can't be closed. Collapse the second rail to icons on a wide screen and every item shows regardless: there's no room for stage names there, so nothing hides behind one.
Picking an agent — in the switcher, or on the Agents list — opens that agent at its first page rather than quietly changing a setting. The list is the front door and the switcher is the shortcut.
Each card says when that agent last answered and how many questions are waiting to be taught, and that agent's own page carries both as well. A Team Agent's card shows only the questions waiting — "last answered" is one conversation's timestamp, and on a small team that comes close to saying who just asked. A card showing a Soon marker is one nothing could read: a figure that failed to load says so rather than showing a zero that would read as never answered.
The Knowledge Hub carries a second level under it — Sources and Add — each with its own address. What it knows and Guardrails are questions about one agent, so they belong to that agent's own Knowledge page rather than to the account's Hub: see Agent knowledge — one agent's view. See also Sources.
A page that isn't there for that agent says so
Not every page exists for every kind of agent: a personal twin has no Escalation, a Team Agent has no Activity. Opening a link to one of those — a bookmark, a shared address, a page you used to have — lands you on Overview, and a line at the top says which page you asked for and which one you got.
The address bar is corrected to match. It used to keep naming the page you asked for while showing you a different one, so copying the link passed the same surprise on to whoever you sent it to. Now the address always names the page actually on screen.
Dashboard says "Dashboard"
The trail at the top of the screen reads Dashboard / Insights on this section, and the Agents list and the Knowledge Hub carry theirs the same way. They used to read the same as an agent's own pages, which made two different pages announce themselves identically — and the one that was wrong was this one, because its figures span every agent you own rather than the one that was being named.
Inside an agent there is no trail at all. The one segment it still carried was the agent's own name, and that name now sits on its own line — small, in italics — directly above the page title. Same fact, moved into the column it is actually about instead of the chrome above it. It is still there on every one of that agent's pages, which matters more than it sounds: collapse the rails to icons and they show a kind, never a name, so that line is what tells you whose Train page you have open. The account pages keep their trail because they belong to no one agent, so there is no name there to bring down.
Nothing on this page reorders itself
It used to: three cards at the bottom were ordered by what you opened most. With the September 2026 split those three became named blocks — one here, one on the agent's page, and one dropped — so there is nothing left for a ranking to arrange, and the order above is the order every time. The rails don't move either. A fixed list you can learn is worth more in a menu than a good guess.
Which agents each figure covers
Dashboard and Knowledge are about your whole account. Overview, Insights and Activity add up every agent you own; Agents is the only section about one at a time.
Because "every agent" isn't always the same set, each block of figures says which agents it covers — Across all 3 of your agents, or Across 2 of your 3 agents, under the numbers. On a computer, rest your pointer on that line to name them.
Two things make the set smaller than your whole account:
- A Team Agent never appears in Activity or in any per-conversation figure — the recent-conversations list, the most-cited tally. It counts in the totals (conversations, questions, people, the activity chart) and in Top questions, because a tally of what was asked names nobody who asked it. See Team Agents.
- An agent shared with you is somebody else's, so none of your figures include it.
Where a figure covers one agent, you can change which one. Beside each of those lines on Insights — the deflection rates, the teach queue, the whole Outcomes block — is that agent's name with a chevron. Open it, pick another, and the page reloads showing that agent's figures; the combined chart and Top questions don't move, because they already covered everything. With one agent there is no picker and no scope line, because there is nothing to change it to.
A few things stay about one agent. Since the split most of them simply live on that agent's page — the teach queue, the setup bar, the readiness reading and the publish state are all on an agent's Overview, where the page names the agent and no caption is needed. What is still one agent's on a combined page says so where it is printed: the deflection and escalation rates and the whole Outcomes block on Insights, and everything under Knowledge. The per-agent usage meters in Plan and usage name their agent on every row, for the same reason.
Since you last looked
That line is on the agent's own page, not here — it names one agent, and every number in it is that agent's.
Answered from your material
On the Overview this figure is labelled Questions that raised nothing to teach. The heading here keeps the older name because other pages link straight to it; the number is the same one.
The same figure heads Insights, in the strip along the top, beside the deflection and escalation rates and the count still waiting in your teach queue. The strip is two cells wide until the rates arrive, and stays two if your plan doesn't include them — it is four only when there are four real figures to put in it. Same label in both places, so the two pages can never look like they're counting different things.
Two counts side by side: questions counted this week, and how many your hiy answered without saying it couldn't. The second is the first minus everything that landed in your teach queue. Every visitor message counts as a question, follow-ups included.
The shortfall isn't a grade — it's your teach queue, and it links straight there. Answer a question once, in your own words, and never answer it again.
What it does not measure. Not whether the answer was right. Not whether the person went away happy. Not where the answer came from. hiy has no honest signal for any of those and won't invent one — the same rule the deflection and escalation rates are held to.
It's counts, never a percentage. You get the two numbers the figure is made of, so you can check it. A percentage off a handful of conversations is false precision dressed as a measurement.
Three known ways it reads high. Repeats don't stack — two people hitting the same wall in a week queue that question once. A decline worded in a way hiy's detector doesn't recognise counts on the good side. And your teach queue stops at 100 open questions; past that a new one isn't written down at all, so the figure climbs while the queue stands still.
The 400-message undercount is gone (September 2026). Both counts used to be folded from a read of your 400 most recent messages, so a busier week counted only part of itself. Both are now exact counts over the seven days, across every agent you own. The three ways it reads high still stand.
If it falls, look at the detector before the agent. This number depends entirely on how well hiy recognises its own agent saying I don't have that. Make that stricter and more questions reach your teach queue, so the figure drops with not one answer having changed. That's the measurement starting to work.
Top questions
On Insights. The questions people repeat, counted across every agent you own — a tally of what was asked names nobody who asked it, so a Team Agent counts here where it counts in no per-conversation figure.
It reads your most recent questions, not all of them. A budget of 400 rows per agent, shared between them, never more than 999 in total — the database answers at most 1,000 rows to one read, and the last is held back to tell whether there were more. A busy account is therefore a tally of recent traffic. When the cap bound, the panel says so: More than 999 questions on this account — the tally counts the 999 most recent. No line means nothing was left out.
Most cited
At the foot of This week. Which of your sources your agents actually lean on — a useful signal for what to make more of: if one post is behind half your answers, that's what your audience wants.
Folded from the per-conversation read rather than the head counts above it, so it covers a smaller set of agents: a Team Agent isn't in it, and the line under the tally says so in its own words rather than borrowing the one over it.
Outcomes: leads and clicks
Insights shows the last 30 days of outcomes: the questions people were asking when they left their contact details, the links they followed out of your answers, and — separately — clicks on booking links. Each opens the conversation behind it; see Where a lead came from.
Deflection and escalation
Insights opens with two rates over the same conversations: how often your agent said it couldn't answer, and how often it never did. Below twenty conversations neither is stated — see Deflection and escalation.
The lists on Insights, and what "Show all" means
Your knowledge gaps, the questions that became leads, and the links people clicked are tables. Press a column heading to sort; press again to reverse. Each shows six rows to start, with Show all underneath when there are more — and sorting reorders what is on screen, so press Show all first if you want the oldest row rather than the oldest of the six newest.
"All" means all the rows this page was given, and no more. Each read has a budget — your teach queue twelve deep, your recent leads fifty — so the number on the button counts what was loaded, not everything that exists. A button offering you every lead over a read that fetched the fifty most recent would be telling you something nobody checked.
An empty list gives you something to press: why it's empty, and the one thing that would change it — add a source, put your agent on your site, add a link to your profile. What it never does is print an empty list because a read failed, because "no bookings yet" is a claim about your week and a broken query is not evidence for it.
These numbers don't arrive by email yet. You can set the summary's schedule and preview exactly what it would say, but nothing is being sent. See Email notifications for the schedule, the preview, and what the summary is made of.