Where
Outreach→AI Agents
Your role needs
Read access to AI Agents
read-campaigns. Admin, Member and Viewer have it by default.To create or change
create-campaigns to add one, update-campaigns to change one, on top of the permission above.If you cannot find this in your sidebar, your workspace may have a custom menu configuration. Contact support and we will check it for you.
Where to look, and what the numbers mean
Open the agent from and stay on its Dashboard tab. The banners sit at the top of it, above the activity feed and the side rail. Each one carries a title, a line explaining it, and the button that deals with it. One card can sit above them, and only one. That is the first-leads card, and it appears for a while on an agent you launched yourself from the creation wizard. The agent built for you at signup never shows it, so on that agent the banners really are the first thing on the tab. When an agent has nothing wrong with it there are no banners at all. The strip is absent rather than empty, so a bare dashboard is a good sign. Two other tabs repeat a subset. The Settings tab raises the missing sending channel and the goal with no call-to-action, because those two are the ones you fix there. The Sources tab raises the two sourcing problems. Everything else appears only on the Dashboard tab.On the agents list, each agent carries a count of its open issues next to its status. That count is worked out from what the list already loaded, and two problems, No leads matched yet and No WhatsApp-reachable leads yet, are judged from figures only the agent’s own page fetches. The list can therefore show a smaller number than the agent does. The Dashboard tab is the one to trust.
The three banners every new agent shows
None of the three is a fault. They are the three things the product deliberately leaves for you, and it raises them in this order: something to send from, the go-ahead to start sending, and somewhere to send people.No sending channel connected
The agent has nothing to send from: no email account, no LinkedIn account and no WhatsApp account. A new workspace has none, which is why this is the first thing a first agent says. Connect senders (1) opens the Add senders overlay on top of the page you are on, rather than sending you to the Email Accounts page. Connecting your first mailbox is a page of its own. An agent set to rotate across every email account in the workspace still shows this banner while the workspace has no email account. Rotating across an empty list sends nothing, so the banner is correct.Outreach is paused while sourcing runs
The agent’s lead sourcing is producing and its outreach is switched off, so leads are arriving and nobody is contacting them. It appears only in that exact combination. A source that is queued and has not started yet does not count as running, so a workspace whose first search is still waiting its turn sees no banner here even though outreach is off. Resume outreach (2) acts where you are: no page change, no dialog. The button shows a spinner while it works, then confirms with Outreach resumed. If it fails, a general error message appears and the banner stays.Goal is missing its call-to-action
The agent knows what it is trying to achieve but has nowhere to send anyone. Your first agent is built with its goal set to Meeting Link and no link attached to it, so this banner is up from the moment the agent exists. Set up goal opens the Goal section of the agent’s Settings tab. There you can give the meeting goal a destination: pick one of your meeting links, or paste a booking URL from any other calendar tool. You can also change the goal to Website Link, WhatsApp or Custom. Every goal has its own field, and the banner clears when that field is filled. That is a link for a meeting, a website address, a phone number, or the description you write yourself for a custom goal.When sourcing stops
Sourcing paused: out of credits
At least one of the agent’s lead sources stopped because the workspace ran out of credits. If you are still on a trial, what your trial credits cover is worth reading before you buy more. Get more credits opens the subscription dialog over the page. It does not navigate anywhere, so you keep your place on the dashboard. Once the workspace has credits again, you do not have to restart anything. That applies whether you bought credits or a new billing period started. We re-check every source paused for this reason on a schedule, and put the ones that can pay back to work. It is not immediate. The source stays paused and the banner stays up until that check next runs, so a source still sitting there a while after a top-up is behaving normally.Sourcing reached its company limit
At least one source has analyzed every company its limit allows. The same banner, with the same words and the same button, also appears when every one of the agent’s searches has finished: a different situation with a different meaning. Two things tell them apart:- The Setup checklist in the side rail counts sourcing as done for a finished search, and leaves it needing attention for a source stopped at its limit. A green checklist row under an amber banner means the searches finished.
- The Sources tab raises this banner only for the limit case. If the banner is on the Dashboard tab and not on the Sources tab, the searches finished.
The other banners a first agent can raise
The missing channel and the goal with no call-to-action are the two the product treats as blocking, and the two that also put a dot on the Settings tab. Nothing in this table is marked that way. Agent alerts and what to do about them covers all of them in full.
When onboarding itself goes wrong
The onboarding wizard is shown in English whatever language the rest of the app is set to.
”We couldn’t analyze that website. Let’s set it up manually.”
The analysis of the website you entered failed. The wizard moves you straight to the first of its four manual questions: what your business is called, what it does, its language, and the industries you sell to. Those are the same four fields the analysis would have filled in, so nothing is lost but the typing. The address you entered is kept. Going back from that first question returns you to the website field, so you can correct the address and try again rather than typing everything out.The analysis is taking too long
The wait screen narrates what it is doing one line at a time. That narration is the same timed script the product runs everywhere it waits on a model. It is not reporting progress, and it never announces that it finished. If the analysis runs long, a panel appears under it headed This is taking longer than expected, offering Keep waiting, Continue manually, and (only if you already belong to a workspace) Back to workspaces. A Change website link sits under the caption the whole time, panel or no panel. Keep waiting only dismisses the panel. The analysis was never stopped and it moves you on by itself if it finishes. The other three stop the app waiting for it.We couldn’t build your first agent
Your first agent is built for you after onboarding, on our side. While that runs, the AI Agents page shows a card headed Generating your first AI agent with a list of stages beside it. Those stages are a timed animation: one lighting up does not mean it finished. If the build fails, the card replaces the stages with We couldn’t build your first agent and a single Try again button, which starts the build again and goes back to waiting. If the full-screen tour that opens over the page is still up, the same message and the same button appear there instead. While the card is waiting, it checks the build’s status on a loop. A check that fails on its own (a dropped connection, a moment of network trouble) is swallowed and tried again. A card still showing stages is therefore not hiding an error from you. Try again does not work that way. If the request that restarts the build cannot be sent at all, the same failure message comes straight back. Press the button again.The AI Agents page is empty and nothing is being built
You land on the AI Agents page after onboarding and find no agents, no card and no error. That means the request that starts the build never landed. Onboarding does not stop for that, it takes you into the workspace anyway, so the failure looks like an empty page. There is nothing to retry from that screen. Build one yourself with Create AI Agent, or use Suggest with AI to have the setup proposed for you, following your first AI agent. The card is also remembered only for the browser tab you finished onboarding in. Open the workspace in another tab, or come back after closing the browser, and the card is gone even though the build may still be running. The list refreshes itself while you watch it, so the agent appears on its own once it is ready.Related
Agent alerts and what to do about them
Every banner an agent can raise, not just the ones a new one shows.
The agent setup checklist explained
The card in the side rail, and why it can disagree with a banner.
No sending channel connected
Every reason an agent sends nothing, in the order worth checking.
No leads found
What to change when sourcing runs and nothing matches.