Monitor running agents in the task feed

When you run an agent, qualify or research records, find contacts, or run a workflow, the work happens in the background. The task feed is where you watch it. Every task shows what it is doing, who started it, how far along it is, and whether it finished, errored, or was skipped. Use it to confirm a run actually started, to track progress on a large batch, and to stop something that is still queued or running.

There are two places to look: a global feed that covers the whole workspace, and a per-record view that shows only the tasks touching one account or contact.

Where the task feed lives

Global feed. In the top header, click Task feed. A panel headed My last agent runs opens with your five most recent tasks - always your own, which is what lets it answer “did the thing I just started actually start” at a glance. Click View all to open the full Agent tasks feed page, where you can filter, page through history, see teammates’ runs, and stop running tasks.

While anything is queued or running, the Task feed control itself animates and swaps its icon for the agent mark, so you can tell from any screen that work is still in flight without opening the panel. When a run finishes it also announces itself as a card at the corner of the screen - see Manage notifications.

Per-record feed. Open an account or contact, go to its Activity tab and stay on Agent tasks. This lists only the tasks for that one record. While work is in progress on the record, an indicator appears on the Activity tab label so you can spot active runs without opening it. Click a task card here to expand it in place and see that record’s own journey through the run’s steps, without opening a separate view.

Both views refresh on their own every five seconds, so progress and new tasks appear without reloading the page.

The header panel is yours alone and cannot be widened. The full Agent tasks feed page also opens on your own tasks, so your view stays focused on your work rather than the whole team’s, but there you can change the Created by filter to bring in a teammate’s runs.

Reading a task

Each task is a card. From left to right it shows:

An icon for the kind of work (a distinct icon marks tasks that came from a workflow).
A plain-language description of what the agent is doing, who it is for (the account or contact), and on finished tasks a running count of results, for example how many accounts came back ICP yes, how many contacts were found, or how many plays were generated.
A line of detail underneath: who started it (on the full feed page - the header panel leaves the name out, since every run in it is yours), how long ago it was created, and what it has cost in credits so far. Credits that have not been drawn yet, on a run that has only just started, simply do not show.
The current status and a progress bar.

Statuses

The status reads as one of:

In queue - the task is accepted and waiting to start.
In progress… - the task is running now. The card shows an animated border while it runs.
Finished - the task completed.
Stopped - the task was cancelled before it finished.
Out of credits - the task stopped partway through because your workspace ran out of credits. Records already completed keep their results. See Resume a run that stopped for credits below.
User limit reached or Team limit reached - the task stopped partway through on a credit limit set for the person who launched it, or for their team, rather than on the workspace balance. Records already completed keep their results, and a top-up does not help - see A run stopped because a credit limit was reached.
Workflow limit reached - the task stopped partway through because it reached the credit limit set on this run itself. Records already completed keep their results; raising or removing that run’s limit continues it - see Resume a run that stopped for credits below.
Error - the task could not be completed.
Skipped - the task did not run for that record, for example because there was nothing to do.

When a task carries extra context about a problem, an information tip appears next to the status. If your ecosystem structure has a logic issue, the tip points you to contact your customer success manager.

The progress bar breaks the batch into Completed, In progress, and In queue so you can see how much of a large run is done.

Filtering the feed

On the full Agent tasks feed page (and the per-record Agent tasks view) you can narrow the list with these filters:

Status - pick one status (In queue, In progress, Finished, Stopped, Out of credits, Error, or Skipped). Status is single-select; the others let you pick more than one value.
Agent steps - the kind of work a task did. The options name the agents themselves, matching the vocabulary used everywhere else in the workspace: Account qualification, Account research, Contact qualification, Contact research, Contact finder, Website finder, Account planning, Play, Waterfalling, and Follow contacts who changed jobs. Two things narrow that list. Only steps that actually appear in the history you are looking at are offered, so it shrinks as your other filters do and you never pick an option that returns nothing. And steps that are not work in their own right are left out: the routing Condition step a workflow uses to send records down different paths, and the internal preparation an agent does for itself before its real work starts. Filtering by those would hand you the run rather than the outcome you were looking for. This filter is available to everyone who can see the feed.
Source - where the task was started from, such as the Data Hub or a workflow.
Created by - the person who started it. Set to you by default; clear it or pick other names to include teammates’ tasks.
Date added - a from/to date range.

Tasks are listed newest first.

Opening the record a task ran on

Hover a task card, in the header panel or on the full feed page, and it grows to reveal its actions. The first is the way into the work itself, labelled with what it will open: Open Acme Corp when the run touched a single record, Open 40 accounts or Open 12 contacts when it touched many. A run that has not completed anything yet offers no such action, because there is nothing to look at.

Where it lands is chosen from the work the run actually did, not from the kind of task it was:

One record opens that record’s page on the tab the last agent wrote to - qualification and research land on the Research tab at the relevant section, account plans on Overview, a contact finder run on that account’s Contacts tab, filtered to the contacts that run touched so a handful of new people are not lost among the hundreds already there. You arrive at the new answer rather than at the top of the record.
Many records open the Accounts or Contacts list filtered to that run, so the batch is together and you can sort it, filter it further, or select it all and run the next agent.
A workflow run follows the record’s own path through the steps, so a record that stopped early lands on what its last completed step produced rather than on what the workflow was configured to end with.

The second action is Analyse with Eva - see the next section. Card actions are on hover on a mouse, on keyboard focus when you tab through the feed, and permanently visible on touch screens.

On the full feed page you can also select several tasks at once and use Open in Accounts or Open in Contacts to view all of their records together.

Seeing the run report

Open the menu on a task card (the three-dot button) and choose one of two views: Run overview for the shape of the whole run, or Per-record breakdown to look at what happened to one specific account or contact.

Analyse with Eva sits on the card’s hover actions beside Open …, and hands that specific run to Eva and asks her what happened and what to do about it. The run comes with the conversation, so she reads it directly rather than asking you which one you mean. Use it when the report tells you what happened - a batch of records skipped, a step that errored, an agent that returned less than you expected - but not why, or when the next move is not obvious from the counts. Both views stay available; she is the faster route to a cause and a recommendation, they are the faster route to the numbers.

Run overview

Whether the run was a single agent or a multi-step workflow, Run overview is a visual map of the steps it used: each step is a card, laid out top to bottom and connected in the order the run executed, so you can see the whole motion at a glance and read how records moved through it.

The flow, drawn out. Every step appears as a card showing the agent’s name and, as records complete, a row of small status counts (hover one to see what it counts - completed, skipped, and so on) together with how long the step took. A branching step fans out into its paths, each labelled with how many records went down it, and a path that ends is marked End of flow. If the run loops a step, each pass shows as its own card, so the repetition is visible rather than hidden. A step no record has reached yet is left off the map entirely, and when two paths lead back to the same next step, that shared continuation is drawn once beneath them instead of being duplicated under each path - so the map reflects what the run has actually done, not every step in the configured setup.
Click a step for detail. Selecting a card opens a panel with three parts. Statistics breaks the step down into how many records completed, are in progress, are pending, were skipped, or hit an error, plus the step’s total time and its average, fastest, and slowest per record. When a step has skipped or errored records, an expandable Why were N records skipped? or Why did N records fail? disclosure lists the specific reasons - for example no research agent assigned to a persona type, or an invalid domain - each with how many records it applied to, so you can see the cause without opening individual records. Results shows the outcomes the agent produced, grouped by kind - for example how many accounts came back ICP yes or no, a breakdown of why contacts were disqualified (out of date at their company, seniority or expertise mismatch, and so on), how many contacts were found per persona, and how many verified emails and phone numbers were returned along with which data vendor supplied each. Switch between All, New this run, and Reused to tell fresh work from outcomes carried over from an earlier run - All is the combined total and is shown by default. Setup is a read-only summary of how the step was configured for this run, such as which verticals it checked against or which plays it wrote; for a workflow’s action steps it lists the tags applied, or the Custom field and the Value written. Those steps are named Tag account, Tag contact, Set account attribute and Set contact attribute in the report.
What the run cost. The total credits the run has drawn so far sits in the report header, and each step card carries the credits that step spent, so an expensive run tells you immediately which part of it was expensive. Selecting a step shows the same figure in its detail panel. If the run has a credit limit, the header figure becomes a small bar reading credits used against the limit (hover it for “N of M credits used”); it turns amber once less than 70% of the limit is left and red below 30%. See Credits and billing.
Change the run’s credit limit. If you can launch workflows, a pencil beside that figure - in Run overview and in Per-record breakdown - opens Workflow credit limit: type an amount and select Save limit, or Remove limit to take the cap off. Selecting it on a run without a limit lets you add one while it runs. The run pauses once its spend reaches the limit. See Cap what a run can spend.
Move around the report. Drag an empty part of the canvas to pan, and use the zoom controls in the bottom-right corner (a minus and a plus, with the current percentage between them; click the percentage to reset it). While a run is still going the report refreshes on its own every few seconds, so the counts fill in live.
Opens on the step you’re waiting on. On a run that’s still going, the report centers itself on the first step that is in progress (or, if nothing is running yet, the first one still queued) rather than the start of the flow - useful on a long run where the early steps already finished and you want to see what’s happening now. This only happens once when you open the report; if you pan elsewhere yourself, later refreshes will not pull the view back.

Per-record breakdown

Choose Per-record breakdown to search the run’s records by name, pick one, and see just its own journey: the steps it went through, in the order it went through them, each shown as a small icon for the agent that ran it, plus a dashed icon for any branching decision along the way and which way it went. Expand a step in that journey to see what it logged for this specific record - contact-finder results, qualification checks, which criteria were researched, and similar per-agent detail. A research step lists its criteria under Research results, split in two when both apply: what this run did (for example “3 researched this run, 1 not found”) and what was already researched by an earlier run and reused rather than paid for again. Each criterion shows Found or Not found, its answer, and whether it was Researched this run or comes From a previous run, with the date. A research step that was skipped researched nothing, so everything it shows counts as already researched. A contact who left their company, in a workflow that looks for replacements, shows a Find replacement step in their own journey: it starts with “Finding another [persona or job title] as [name] left [account].”, lists what each search turned up, and ends either with the replacement or with “No replacement found at [account] after [n] searches.” A Waiting line in a contact-finding step means another search on the same account had to finish first. Each record in the list also shows what it cost in credits, which is the fastest way to see whether a heavy run was heavy across the board or driven by a handful of records that needed the most work. When the run only ever covered one record, this view skips straight to that record’s journey instead of asking you to search for it. Use this when you already know which account or contact you care about and want its story directly, rather than reading it out of the whole run’s aggregate statistics.

Retry the records that errored

When a run has failures, a red banner sits at the top of the report summarising them: how many records failed, with what error message, on which step, and the first few records by name. Two actions sit on it.

Show failed narrows the per-record list to just those records, so you can read what happened on each one.
Retry errors opens Retry errored entries?, which lists each group of failures with its step, its count, and its error message, and confirms with Retry N entries. Accepting resets those steps to pending and re-queues those records; everything that already succeeded is left alone. A message confirms that the errored entries have been queued.

Retrying re-runs work, so it draws credits again for the records it re-queues. For that reason the button only appears for people whose team is allowed to launch workflows - the same permission behind Continue workflow below. If you cannot see it, you can still open Show failed to diagnose, and ask someone who can launch runs to retry.

This is the right first move when the failures were transient - a data provider that did not answer, a step that timed out. When the cause is missing input on the records themselves, fix the records first, or the retry fails the same way. See My agent job failed.

Resume a run that stopped for credits

If a run stops with the Out of credits status, a Continue workflow button appears - on the task card itself, and inside both Run overview and Per-record breakdown - so you don’t have to start the batch over. Clicking it picks the run back up from where it paused; records already completed are not redone.

The button is available to anyone with permission to launch workflows. It stays disabled, with a tooltip explaining why, until your workspace has credits again - use the Top-up link shown next to the message to reach the top-up page, or see Top up your credits. Once the workspace has credits, click Continue workflow to resume. A run labelled User limit reached or Team limit reached resumes the same way, but only once that limit has been raised or has reset - see A run stopped because a credit limit was reached.

A run labelled Workflow limit reached stopped on its own per-run limit, so there is no Continue workflow button and no Top-up link. Instead:

Select Raise limit & continue in Run overview or Per-record breakdown, or the resume icon on the task card (its tooltip reads “Raise or remove the workflow limit to continue”).
The Workflow credit limit box opens, showing how many credits the run has used so far and pre-filled with an amount just above that. Type the new limit - it has to be above what the run has already used, or the box says “Must be above the N credits already used.” - and select Raise limit & continue. Leave room for the work still to do: each step needs space for its full possible price before it starts, so a limit only just above what was used can pause the run again straight away.
Or select Remove limit & continue to let the run finish with no cap.

Either way the run picks up where it paused, and you see Workflow resumed successfully. A contact search that was interrupted by the limit looks only for the contacts it still had to find, so resuming never buys the same contacts twice. Both need the permission to launch workflows.

Stop a running task

You can stop a task that is In queue or In progress. Stopping is available from the full Agent tasks feed page, not from the per-record Agent tasks view. You can stop tasks you started yourself; full-access members can stop any task.

On the Agent tasks feed page, find the task you want to stop.
Open its menu (the three-dot button on the right of the card) and choose Stop.
A Stop task? confirmation appears, naming the task. Click Stop to confirm.

You will see a Task stopped successfully message, and the task moves to the Stopped status. Any records already finished before you stopped keep their results; the remaining queued records are not processed.

What happens next

New tasks appear in the feed within a few seconds of starting a run. A task moves through In queue to In progress to Finished as the agent works through the batch; the progress bar fills as records complete. The time a run takes depends on the kind of work and how many records are in it.

When a task reaches Finished, its results are written to the relevant records. Click the card to open those records, or open Run overview for the step-by-step breakdown, or Per-record breakdown to check one specific record. If a task lands on Error, see Fix an agent job error. If a record shows as Skipped, see A record was skipped. If a task stops as Out of credits or Workflow limit reached, see Resume a run that stopped for credits above.