Read the Runs panel
Every run of an agent, step by step. This guide covers what each status means, how to read a step and its output, approving and stopping runs, the messages a failed run shows and what to do about each, and who can see runs.
Checked against the app on
Open the Runs panel
The Runs panel lists every run of the agent you have open in Agent Builder: when it started, how long it took, what it cost and whether it worked. Select a run to see each step in the order it ran. It is also where you start a run by hand with Run now, and where you approve or reject a run that is waiting at an Approval step.
Build modular AI workflows with reusable nodes
Weekday mornings
Daily at 09:15 · Mon, Tue, Wed, Thu, Fri
Fetch unpaid invoices
List invoices
One item per invoice
From invoices
Overdue by 1 to 14 days
Keep ALL of 3 rule(s)
Each overdue invoice
For each of incoming items
Over £1,000?
1 rule
Draft chase email
1 tool connected
Check before sending
Approve in panel
Send chase email
1 action enabled
Runs
Each run executes the workflow node-by-node. Select a run to follow its steps.
Triggered
list_invoices executed
Extracted 12 item(s), 0 empty, 0 invalid
Filtered 3 kept / 9 discarded
Loop item 1/3
Condition routed 1 true / 0 false (per item)
AI step complete (1 item(s))
- 1The agent's totals since it was created: how many times it has run and the automation credits it has used.
- 2Run now, to start a run yourself. Refresh, beside it, reloads the list.
- 3The run list, newest first.
- 4The run you selected: a header with its status, time and credits, then every step it ran.
- 5Minimise and Close.
To open it, click Runs in the builder's toolbar. The button appears once the agent has been saved, so a new agent shows it after your first save. Starting a run from Run now opens the panel for you and selects the new run a moment later.
Weekday mornings
Daily at 09:15 · Mon, Tue, Wed, Thu, Fri
Fetch unpaid invoices
List invoices
One item per invoice
From invoices
Overdue by 1 to 14 days
Keep ALL of 3 rule(s)
Each overdue invoice
For each of incoming items
Over £1,000?
1 rule
Draft chase email
1 tool connected
Check before sending
Approve in panel
Send chase email
1 action enabled
- 1Runs, in the toolbar.
- 2The message after you start a run. A dry run says Dry run started. No real actions will be performed. See the Runs panel. instead.
Minimise (the line at the top right) hides the panel and keeps your place, so the same run is still selected when you open it again. Close (the cross) clears the selection. The list keeps itself up to date while the panel is open, so you can leave it open and watch a run progress.
The top of the panel
Beside the one-line description at the top are the agent's totals and the buttons that act on the whole list.
Runs
Each run executes the workflow node-by-node. Select a run to follow its steps.
Select a run to see its step timeline.
- 1The agent's totals since it was created, such as runs: 132 · credits consumed: 1,284.6.
- 2Auto-refreshing. The list reloads every 4 seconds while any run is in progress, and every 12 seconds otherwise. The duration of a run in progress counts up every second.
- 3Refresh reloads the list straight away.
- 4Run now opens the Run workflow window, where you choose a full run or a dry run. It reads Running while the run starts.
- 5Minimise and Close.
- If the canvas has a Spike Test step, a Run spike test button appears after Run now.
- People with view-only access do not see Run now or Run spike test. Who can do what is covered under Who can see and act on runs, below.
- On an agent that answers phone calls, the panel lists calls instead of runs and the totals count calls, such as 14 calls · credits consumed: 9.8. Make outbound calls and answer incoming calls covers that view.
How to start a test, and the difference between a full run and a dry run, is in Test an agent before it goes live.
Read the run list
Each run is a card with three lines. The first is its status, with a Dry run or Spike badge where that applies. The second is when it started, in UK time, and how long it took. The third is what started it and the automation credits it has used.
- Queued: the run has been created but has not started its first step.
- Running: steps are in progress, and the duration is blue and counts up. A run waiting for a slow step to finish in the background, such as Browser Use, also shows Running.
- Awaiting approval: the run is paused at an Approval step until someone approves or rejects it.
- Success: every step that ran finished without an error.
- Success with issues (1): the run finished, but the number in brackets is how many steps failed along the way. This happens when a failing step is set to carry on. Open the run and look for the step marked Failed.
- Failed: a step failed and stopped the run, or the run could not start. The reason is shown at the top of the run.
- Cancelled: someone stopped the run with Stop run.
The trigger is shown as the app stores it: Schedule, Manual for anything started with Run now (full and dry runs alike), Chat, Email, Spike for a spike test, and the system's name for a system alert, such as Fixflo, Simpro or Bigchange_rest for a BigChange webhook. Durations read like 850ms, 4.1s, 38s, 2m 41s or 1h 04m, with a dash until the run starts. The credits figure grows as the run goes and is final once it ends.
The list shows 25 runs at a time, newest first. Once an agent has run more than 25 times, a line such as Showing 1-25 of 132 appears under the list, with Previous and Next to move through it.
Runs
Each run executes the workflow node-by-node. Select a run to follow its steps.
Showing 26-50 of 132
1Triggered
list_invoices executed
Extracted 12 item(s), 0 empty, 0 invalid
Filtered 3 kept / 9 discarded
- 1Which runs you are looking at, out of the agent's total.
- 2Previous and Next move 25 runs at a time.
There is no search, and no filter by date or status, in the panel. To find a run from a particular day, ask the AI builder (see Other places to check runs, below). Runs stay with the agent: there is no setting to clear them and no way to delete a single run.
Open a run
Click a run to open it on the right. The header gives its status, how long it took, how many steps it recorded, the credits it used and when it started. Underneath, every step is listed in the order it ran, numbered from 1.
Triggered
list_invoices executed
7Extracted 12 item(s), 0 empty, 0 invalid
Filtered 3 kept / 9 discarded
Loop item 1/3
Condition routed 1 true / 0 false (per item)
AI step complete (1 item(s))
- 1The run's status, with any Dry run or Spike badge beside it.
- 2How long the run took from start to finish. Time spent waiting for an approval counts.
- 3How many steps the run recorded. A step inside a loop counts once for each item.
- 4The credits charged for the whole run. Hover over it for Total Agent Builder automation credits charged for this run, across every step and tool call.
- 5When the run started, in UK time.
- 6One step: its number, its step type, how long it took and whether it worked.
- 7A one-line summary of what the step did. Click the step to open it.
Steps are named by their type, such as BigChange REST or Intelligence, not by the name you gave them on the canvas. When an agent has two steps of the same type, tell them apart by their order and summaries. Until the first step starts, the run shows No steps recorded yet.
The panel shows credits for the whole run, not for each step. To see what each kind of step costs, see Automation credits and usage limits.
Open a step
Click a step to open it. The grey line at the top gives when it started, how long it took, and how many items left by each of its outputs, such as Output: main ×3, discarded ×9 for a Filter that kept 3 invoices and set 9 aside. A step that was retried also shows which try this was, such as Attempt 2/3. Under that are the summary and Step output: the data the step passed on, as raw data (JSON).
Extracted 12 item(s), 0 empty, 0 invalid
Filtered 3 kept / 9 discarded
Step output
{
"handles": {
"main": 3,
"discarded": 9
},
"items": {
"main": [
{
"invoiceNumber": "INV-20418",
"contactName": "Northgate Housing",
"siteName": "Riverside House",
"amountDue": 1240,
"dueDate": "2026-09-30",
"daysOverdue": 9
},
{
"invoiceNumber": "INV-20433",Extracted 184 item(s), 0 empty, 0 invalid
Step output
{
"handles": { "main": 184 },
"items": {
"main": [
{ "jobRef": "J-48213", "site": "Harbour View Apartments", "status": "Completed" },Showing the first 50 items per output.
3Extracted 2410 item(s), 0 empty, 0 invalid
Step output
{
"handles": { "main": 2410 },
"itemsOffloaded": true,
"itemCount": 2410,
"summary": "Extracted 2410 item(s), 0 empty, 0 invalid"
}Output too large to show here (2,410 items). Downstream steps still received it in full.
4- 1When the step started, how long it took, and how many items left by each output: main for the 3 invoices kept, discarded for the 9 set aside.
- 2Step output: what the step passed on, item by item. Long output scrolls inside the box.
- 3When an output has more than 50 items, the panel shows the first 50 of each and says so.
- 4When the output is very large, the panel shows only the counts and summary. The next steps still received every item.
- The panel shows what each step passed on, not what it received. To see what a step was given, open the step before it.
- A step that passed nothing on and has no summary shows No output recorded.
- Files in the output show as a short note of their type and size rather than the file itself.
- Summaries are written by the app and can be technical, such as list_invoices executed. They are most useful for counts: how many items were found, kept or sent.
Loops, branches and tool calls
Some steps appear several times in one run, and some you might expect never appear at all. This is the rest of the 09:15 run: three invoices went through the Loop, and only the two over £1,000 were written up and sent.
Loop item 1/3
1Condition routed 1 true / 0 false (per item)
AI step complete (1 item(s))
file_search executed
Approved
send_email executed
Loop item 2/3
Condition routed 0 true / 1 false (per item)
3Loop item 3/3
Condition routed 1 true / 0 false (per item)
AI step complete (1 item(s))
file_search executed
Approved
send_email executed
Loop complete: 3 item(s)
4- 1A Loop records a line each time it starts on the next item, such as Loop item 1/3.
- 2tool marks an action the AI step took itself with a tool connected to it, here a search of your files. Tool calls appear under their technical name, such as core.file_search.
- 3This Condition sent the second invoice down its false path. Nothing was connected there, so the next line is the Loop moving on to the third.
- 4Once every item has been through, the Loop records Loop complete: 3 item(s).
- Loops: the steps inside a Loop appear once for each item, between its Loop item lines, so a loop over 12 invoices makes a long list. A Loop that works through a large dataset in batches shows lines such as Dataset batch 1-50 of 200 instead.
- Branches: a step on a path that received nothing is not shown at all. The run is not missing a step; that path had nothing to do. The summary of a Condition, such as Condition routed 0 true / 1 false (per item), tells you which way items went.
- Spike Test and On run error: Spike Test steps run only in a spike test, and an On run error step runs only when the run fails, so neither appears in a normal run.
- Tool calls: each action an Intelligence step takes with a connected tool gets its own line, marked tool, just after that Intelligence step.
Conditions, loops and approvals explains how Loops and Conditions decide where items go.
Runs that are waiting
A run can pause part-way, and what the panel shows depends on why. At an Approval step, the top of the run reads This run is awaiting approval. with Approve and Reject underneath. A run waiting for a slow step to finish in the background shows a grey note instead, and needs nothing from you.
This run is awaiting approval.
Triggered
list_invoices executed
Extracted 12 item(s), 0 empty, 0 invalid
Filtered 3 kept / 9 discarded
Loop item 1/3
Running Browser Use in the background. This can take a moment.
Triggered
Started browser task (goal: "Log in to the Harbour Supplies portal and download the September statement")
Browser activity
RunningOpened the supplier portal sign-in page
4.2sportal.harboursupplies.co.uk/login
Signed in as accounts@meridianfm.co.uk
11.8s
Opened Statements and chose September 2026
26.4sportal.harboursupplies.co.uk/statements
- 1Approve lets the run carry on down the Approval step's Approved path.
- 2Reject sends it down the Rejected path.
- 3A run waiting for a slow step, here Browser Use, shows this note. It carries on by itself.
- 4Browser activity: what the browser has done so far, and the browser minutes used. It updates every couple of seconds.
After Approve, a message reads Approved, resuming run. After Reject, it reads Rejected, stopping run, but the run then carries on down the Rejected path if anything is connected to it. An approval given by email reply resumes the run in the same way. Conditions, loops and approvals covers setting up approvals and their timeouts.
Every run must finish within 8 hours of starting, waiting included. An approval left longer than that fails the run with Workflow run exceeded max wall-clock time (8 hours) when someone does answer it.
An integration connected to an Intelligence step as a tool has its own Pause for approval setting. When it pauses a run, the panel shows the run as Running and has no Approve or Reject buttons for it, so nobody can approve it from here. Until that changes, put an Approval step in front of the action instead.
When a run fails
When a step fails and stops the run, the run shows Failed and the reason appears in red at the top of the run. Open the step marked Failed to see the same message, and, if the step was set to retry, how many tries it had. A step that ran out of time is not retried.
Triggered
No output recorded.
UnauthorizedException BigChange REST is not connected
3file_search executed
Approved
No output recorded.
NodeAbortedError Node timed out
5Loop item 2/3
- 1Why the run stopped.
- 2Attempt 3/3: the step was set to try three times, and the third try failed too.
- 3The error from the step that failed.
- 4Success with issues (1): the run finished, but one step failed along the way.
- 5That step. It was set to carry on after an error, so the run moved on to the next invoice.
Some messages start with a technical word, such as UnauthorizedException or NodeAbortedError, and some join several causes with a ← arrow. Read past the first word: the plain-English part that follows is the reason.
Whether a failed step stops the run is up to you. In the step's Advanced: error & retry settings, Stop the workflow fails the run, while Continue (pass input through) and Continue via error branch carry on and the run ends as Success with issues. The same settings let a step retry, and an On run error step can tell someone when a run fails. Centrus does not alert anyone about a failed run by itself, so add an On run error step to any agent you rely on. Conditions, loops and approvals explains these settings.
These are the messages you are most likely to see at the top of a failed run, word for word, with what to do about each.
This organisation has used its 10000-credit monthly Agent Builder allowance. The allowance resets on 01/11/2026.
This run has reached its per-run automation-credit budget (250 credits). Raise the workflow's budget in its settings to let it continue.
This organisation has reached its limit of 10 concurrent workflow run(s).
UnauthorizedException BigChange REST is not connected
NodeAbortedError Node timed out
Workflow run exceeded max wall-clock time (8 hours)
Workflow run exceeded max step count (possible loop)
Run cancelled by user
- This organisation has used its 10000-credit monthly Agent Builder allowance. The allowance resets on 01/11/2026. The organisation's automation credits have run out for the month. No agent can start a run until the date shown, unless an admin adds credits. See Automation credits and usage limits.
- This run has reached its per-run automation-credit budget (250 credits). Raise the workflow's budget in its settings to let it continue. The run stopped part-way at its own spending limit. The limit is 250 credits unless someone has changed it, and can go up to 1,000. Raise it in the agent's settings if the run genuinely needs more.
- This organisation has reached its limit of 10 concurrent workflow run(s). Ten runs across the organisation were already in progress or waiting for approval. The new run fails rather than waiting its turn, so start it again later, and approve or stop any runs left waiting.
- This organisation does not have an active Agent Builder Professional subscription. The organisation's plan does not include Agent Builder at the moment. Ask an admin to check the plan.
- UnauthorizedException BigChange REST is not connected, Simpro is not connected to this agent, Stored Simpro credentials are invalid. Please reconnect. and Outlook not connected to this agent all mean the account a step uses has been disconnected or its sign-in has expired. Reconnect it in that step's Connection section, then run again.
- NodeAbortedError Node timed out: one step ran past its time limit, and timeouts are not retried. Give the step less to do in one go, for example by filtering out items earlier in the agent.
- Workflow run exceeded max wall-clock time (8 hours): the run passed 8 hours from its start, including time spent waiting for approval.
- Workflow run exceeded max step count (possible loop): the run passed 10,000 steps, which usually means items are going round in a circle. Check the wires into your Loop.
- Run cancelled by user: someone with edit access clicked Stop run.
If Run now cannot start a run at all, a message appears at the top of the screen instead and nothing is added to the list. On an agent that has never been saved, it reads Save your workflow as a draft first. That lets you connect integrations and test run it. Otherwise you may see Failed to start run: click Validate in the toolbar, fix any errors it finds, then try again.
Stop a run, or run it again
While a run is Queued, Running or Awaiting approval, its header has a Stop run button. Click it and the run stops at the step it has reached, shows Cancelled, and the top of the run reads Run cancelled by user. Steps that already finished are not undone: an email that has gone stays sent.
Filtered 3 kept / 9 discarded
Loop item 1/3
Condition routed 1 true / 0 false (per item)
Loop item 1/3
Condition routed 1 true / 0 false (per item)
- 1Stop run, in the header of a run that is still going.
- 2The run after it was stopped.
- 3The step that was in progress can stay marked Running, and its time keeps counting. The run itself has stopped; this line is out of date.
Stopping a run that is waiting for approval frees its place among the organisation's 10 runs at a time. Credits used up to the point it stopped are still charged.
There is no button to repeat a past run. To try again after fixing a problem, click Run now. The new run starts from the beginning with fresh data rather than replaying the old one: a scheduled agent fetches today's records, and an Email received agent uses the latest email in its folder. Any step that sent or wrote something in the first run will do it again in the second, so check what already happened before you run again, or use a dry run.
Calls and texts in a run
When a run places a phone call, open the Outbound call step to see the number rung, play the recording and read the transcript. A Send SMS step's summary gives the last four digits of the number it texted and how many message parts it used.
Triggered
list_jobs executed
Outbound call finished
Call · +447700900731
Call recording
Kept until 08/11/2026
Assistant: Hello, this is Centrus calling for Meridian Facilities about tomorrow’s visit to Riverside House. Is now a good time?
Caller: Yes, go ahead.
Assistant: Dylan Hughes is booked tomorrow at 09:30. Can you still make that time?
Sent SMS …0731 (1 segment)
2- 1The call: the number rung, the recording and what was said. A step that did not place a call says No call was placed on this step.
- 2The text, sent to the number ending 0731 in one message part.
In a dry run, nothing is rung or sent, and the summaries say so. An agent that answers its own phone number lists calls rather than runs. See Make outbound calls and answer incoming calls and Send text messages.
Other places to check runs
The Runs panel is the full record, but two other places give a quicker answer.
Agents (2)
Chases invoices 1 to 14 days overdue, every weekday at 09:15
Emails yesterday's completed jobs to the operations team
AI Agent Builder
Edits apply live on the canvas
John Martin · 10:02
Centrus Agent Builder · 10:02
- 1The Usage column of the Agents list: the same totals as the top of the Runs panel.
- 2The AI builder can read the agent's runs and explain what went wrong.
- The Agents list shows each agent's totals in its Usage column, such as 132 runs · 1,284.6 credits.
- The AI builder, inside the builder, can look up runs for you. Ask it to check the last run, or give it a date such as 07/10/2026 to find an older one. It can also see a sample of what each step received, which the panel does not show.
- Usage reports: admins can see automation credits across every agent in the Agent Builder usage report. See Automation credits and usage limits.
Who can see and act on runs
Anyone who can open an agent can see all of its runs, including every step's output. What they can do with a run depends on their access.
- Edit access: see every run, use Run now, Approve or Reject a waiting run, and Stop run.
- View access: see every run and its steps, and use Refresh. Run now, Run spike test and Stop run are hidden, and a run waiting for approval says You have view-only access to this workflow, so you can't approve or reject this run.
Runs
Each run executes the workflow node-by-node. Select a run to follow its steps.
This run is awaiting approval.
You have view-only access to this workflow, so you can't approve or reject this run.
Triggered
list_invoices executed
Extracted 12 item(s), 0 empty, 0 invalid
Filtered 3 kept / 9 discarded
- 1Refresh is the only button. There is no Run now.
- 2A waiting run with no Stop run button beside its status.
- 3The view-only note in place of Approve and Reject.
Step output can include customer names, amounts and the text of emails, and anyone you share an agent with can read it. Share an agent explains how to give someone view or edit access.
On a phone
On a phone, Runs is an icon in the builder's toolbar. The panel opens at almost the full width of the screen but keeps its two columns: the run list takes nearly all the width, and the selected run is squeezed into a narrow strip on the right. The list is enough to see whether today's run worked. To read steps, or to approve or reject a run, use a tablet or computer, or set the Approval step to ask by email.
Centrus
Build modular AI workflows with reusable nodes
Runs
Each run executes the workflow node-by-node. Select a run to follow its steps.
runs: 132 · credits consumed: 1,284.6- 1The description and totals, squeezed into narrow columns.
- 2Run now, partly off the edge of the screen.
- 3The run list, with status, time, duration, trigger and credits for each run.
- 4The selected run, too narrow to read.
More in Agent Builder
- Open Agent Builder
- Build your first agent from a message
- Build and edit agents from chat and WhatsApp
- Find your way round the canvas
- Choose how an agent starts
- Run an agent on a schedule
- Start an agent when an email arrives
- Start an agent from a system alert
- Run an agent from chat and WhatsApp
- Steps an agent can use
- Conditions, loops and approvals
- Send text messages
- Make outbound calls and answer incoming calls
- Test an agent before it goes live
- Publish, pause and restore versions
- Share an agent
- Automation credits and usage limits