Automation credits and usage limits

Each time an agent runs, Centrus charges automation credits for the work it does: 1 credit to start the run, then a set amount for each step that does something, such as reading a BigChange job, asking the AI to write an email or sending a text. Small Team and Large Team both include 10,000 automation credits a month, shared by everyone in the organisation, and an admin can buy more.

Chase overdue invoices

7,846 / 10,000 credits1

Runs3

Settings4

Update & Publish

This organisation has used 7,846 of its 10,000 monthly automation credits (78%).

2

Weekday mornings

Daily at 08:00 · Mon, Tue, Wed, Thu, Fri

List overdue invoices

List invoices · overdue

Each overdue invoice

For each of incoming items

Format amount and due date

2 transform ops

Write chase email

GPT-5.6 Terra

Send chase email

1 action enabled

AI builderMap

Ready to publish

1. 1The credit count: what the whole organisation has used this billing period, out of its allowance. It turns amber at 75% and red at 100%. Rest the pointer on it for the reset date.
2. 2The warning strip appears from 75% of the allowance. See Warnings as the allowance runs low.
3. 3Runs lists every run of this agent with the credits it used. See See where credits go.
4. 4Settings holds the agent's per-run budget and its cost estimate.

Automation credits are separate from message credits. Message credits pay for answers in Chat, on the web, WhatsApp and Teams. Automation credits pay for agents, and an agent's steps never use message credits. See Credits and browser minutes.

A step is charged when it finishes. A step that fails is not charged, but the steps that finished before it in the same run are.

• Every run costs 1 credit, whether it starts on a schedule, from an email, from a system alert, from chat or from Run now. A run that is turned away before it starts, for example because the allowance is used up, costs nothing.
• Integration steps cost 1 credit each time they run, however many records they read or write. A step that lists 200 BigChange jobs costs the same as one that reads a single job.
• AI and knowledge steps are measured: the more text a step reads and writes, the more it costs, with a minimum of 2 credits. The model counts too. GPT-5.6 Luna costs less than GPT-5.6 Terra, GPT-5.6 Sol about twice as much as Terra, and GPT-6 Astra three to four times as much.
• Logic and data steps cost 0.1 each time they run, so a Condition inside a loop of 50 items adds 5 credits.
• Loop itself is free, but every step inside it is charged on every pass.
• Approval costs 0.1 once a decision is made. The wait is free, however long it lasts.

When an Intelligence step uses a tool, such as looking up a job in BigChange or searching your files, each tool call is charged on its own: 1 credit for a lookup in a connected system, and at least 2 for a file or web search. An AI step that makes six lookups adds 6 credits to its own cost.

Three kinds of step are priced by how much they are used rather than at a flat rate.

• Send SMS costs 6 credits for each segment. A plain text of up to about 160 characters is one segment; longer texts, or texts with emoji, are split into more, and each one is charged. Before each text, Centrus checks that it would keep the run within its per-run budget.
• Phone calls, whether your agent makes them with Outbound call or answers them on your Centrus number, cost 1 credit for each started minute once the call connects. Each lookup the Voice AI makes adds 0.1, and recording the call adds 1 when the recording is saved. A call that reaches voicemail is charged like any other connected call. See Make outbound calls and answer incoming calls.
• Browser Use costs 1 credit for each session that starts, plus credits for the work the browser does, measured like an AI step. A session that starts and then fails is still charged. Each session also uses browser minutes: every started minute counts, from an allowance of 300 minutes a month.

Phone calls are charged to the agent rather than to a run, so they do not count towards a run's total or its per-run budget. Each call's credits are listed in the agent's Runs panel and in the Agent Builder Usage report.

Here is one morning's run of an agent that chases overdue invoices in Xero. It starts at 08:00 on weekdays, finds 12 overdue invoices, and writes and sends a reminder from Outlook for each one.

Most of the cost is the AI writing 12 emails. An AI step's cost depends on how much it reads and writes, so the same agent costs a little more or less each day: this one used 51.2 credits on one morning and 47.8 the day before. At around 50 credits a run, 22 runs use about 1,130 of the month's 10,000 credits.

Agent Builder is included with Small Team and Large Team, and both plans come with the same allowance: 10,000 automation credits and 300 browser minutes a month, and up to 100 published agents. The allowance belongs to the organisation, so every agent and everyone who builds one draws on it.

Only admins can open Settings, Billing. The Current Plan card at the top of the tab shows message credits, storage and users, but not automation credits: to see how many you have used, look at the credit count in the builder's toolbar or the Agent Builder Usage report. Roles are explained in Users and roles.

An organisation on any other plan cannot run agents. The builder's toolbar reads Not entitled, and every new run fails with "This organisation does not have an active Agent Builder Professional subscription." An organisation still on an older Small Team or Large Team price sees an Agent Builder Professional add-on on the Billing tab instead, at £200.00 a month, which gives the same allowance.

When your agents need more than the plan includes, an admin can add automation credits in blocks of 5,000 for £50.00 a month, and browser minutes in blocks of 300 for £20.00 a month.

1. 1Click Settings in the sidebar, then the Billing tab. It opens on General.
2. 2Scroll down to Add-ons.
3. 3Click + on Extra Automation Credits once for each block of 5,000 credits, or on Extra Browser Use Minutes once for each block of 300 minutes. The box shows your choice, such as 2 × 5,000 credits.
4. 4Click Update add-ons. Processing addons appears, then Redirecting to payment gateway, and the payment page opens.
5. 5Pay on that page. The extra credits are added to your allowance once the payment goes through.

On a phone, the tiles stack one under another and work the same way.

09:41

Centrus

Add-ons

Extend your plan with additional resources

Meeting Agent

£19.99 per month

1,000 minutes + 1 bot

-0+

Extra Automation Credits

£50.00 per 5,000 credits, per month

-1 × 5,000 credits+

Tap + for each 5,000

Extra Browser Use Minutes

£20.00 per 300 minutes, per month

-0+

Update add-ons

Add-ons are monthly subscriptions, not one-off top-ups. Each block raises the allowance for every billing period from the day you buy it, so choose what you expect to need each month. The boxes count only the blocks you are adding now, so they start at 0 even when you already pay for some.

For an organisation on any plan other than Small Team or Large Team, both boxes are greyed out with the note Included with Small Team and Large Team. See Add-ons for the other add-ons on this card.

The allowance is monthly, counted from your organisation's billing date rather than the 1st. The count goes back to zero on the same date each month, on monthly and annual plans alike.

• Unused credits do not carry over. Each period starts again with the full allowance.
• Browser minutes reset on the same day.
• Changing plan part-way through a period does not reset the count.

To find the date, rest the pointer on the credit count in the builder's toolbar, which ends with Resets and the date. An admin can also open Settings, Billing, click Subscription and read Current period.

Alongside the monthly allowance, these limits apply to every organisation with Agent Builder.

• 10 runs at once. Runs waiting for an approval or a browser session count too. An eleventh run does not wait its turn: it fails at once with "This organisation has reached its limit of 10 concurrent workflow run(s)."
• 100 published agents. Publishing another is refused with "This organisation has reached its limit of 100 active automation(s). Pause or archive one before publishing another." Drafts and paused agents do not count, and publishing a new version of an agent that is already published is always allowed.
• 8 hours per run, counted from the start. Time spent waiting for an approval counts, so an Approval step needs an answer within 8 hours of the run starting. A run that goes over fails with "Workflow run exceeded max wall-clock time (8 hours)".
• 10,000 steps per run, counting every pass round a loop. A run that goes over fails with "Workflow run exceeded max step count (possible loop)".
• Each step has a time limit: 2 minutes for most steps and 15 minutes for AI steps.
• Phone calls: 3 at once for the organisation, and each call ends at 30 minutes.
• Browser Use sessions stop after 10 minutes and 20 browser steps unless you allow more in the step, up to 30 minutes and 100 steps.

Browser minutes have their own meter. Select a Browser Use step on the canvas and the top of its Details panel shows Monthly browser usage: minutes used out of the allowance, how many remain, sessions running or queued, and the reset date. The bar turns amber at 90% and red at 100%.

Monday mornings

Weekly on Mon at 07:30

Download supplier price list

1. Go to the supplier portal and sign in. 2. Open Price lists…

Save to Price lists

Upload file

1. 1Monthly browser usage is the organisation's total, not this step's. Every Browser Use step shows the same figures.

Every run has a ceiling on the credits it can use, so a mistake such as a loop over far more items than expected cannot drain the month's allowance. The ceiling is 250 credits unless you set another figure for the agent, up to 1,000.

The agent's owner and admins can change it. Other people do not see the field. The Settings button appears once the agent has been saved, and on a narrow screen it shows only its cog icon.

1. 1Open the agent in the builder and click Validate in the toolbar, so the estimate reflects the current steps.
2. 2Click Settings. Agent settings opens on the Agent tab.
3. 3Scroll to Per-run credit budget and read the estimate under it.
4. 4Type the most credits one run may use, from 1 to 1000, then click Save beside the field. Per-run credit budget set to 150. confirms it.

• Read the estimate as a range. It counts each AI step at between its minimum and three times its minimum, and every step inside a Loop between 1 and 30 times. It does not include Browser Use, and it shows Validate to refresh estimate. until the agent has been validated.
• Agents started by an email, a system alert or a call show Idle: 0. Per matching event: and a figure instead, because waiting costs nothing and each event is a separate run.
• Set the budget a little above a busy day. The invoice agent above usually costs about 50 credits and about 125 on a day with 30 overdue invoices, so 150 lets every normal run finish and stops one that goes wrong.
• A figure above 1000 is saved as 1000.
• When a run reaches its budget, it stops before its next charged step and fails with "This run has reached its per-run automation-credit budget (150 credits). Raise the workflow's budget in its settings to let it continue." Steps that finished before it keep their results and their charges.
• Phone calls are not counted towards a run's budget, because they are charged to the agent.

Everyone who opens the builder sees the organisation's credit count in the toolbar. It updates every 30 seconds and changes colour as the allowance runs low, and a strip under the toolbar spells out how much has gone.

Rest the pointer on the count for the detail behind it: credits used, published agents out of 100, runs going and waiting out of the 10 allowed at once, and the date the allowance resets.

• The warnings appear only in the builder. Centrus does not send an email or a notification as the allowance runs low, so check the count or the report if your agents run unattended.
• An organisation without Agent Builder sees Not entitled in place of the count, and a red strip reading "This organisation does not have Agent Builder. It is included with Small Team and Large Team. New automation runs will fail to start until this organisation is on one of those plans."

On a phone, the toolbar has no room for the count, but the strip still appears above the canvas.

09:41

Centrus

Agents

Build modular AI workflows with reusable nodes

Chase overdue invoices

Update & Publish

This organisation has used 7,846 of its 10,000 monthly automation credits (78%).

Still shows on a phone

Weekday mornings

Daily at 08:00 · Mon, Tue, Wed, Thu, Fri

List overdue invoices

List invoices · overdue

Each overdue invoice

For each of incoming items

Format amount and due date

2 transform ops

Write chase email

GPT-5.6 Terra

Send chase email

1 action enabled

Ready to publish

Once the allowance is used up, new runs fail to start until it resets or an admin buys more. Agents stay published, but a run that could not start is not tried again later: it is listed as failed in the agent's Runs panel, with the reason.

• A run that has already started carries on and can finish, even if it takes the organisation past the allowance.
• Building carries on. You can keep editing agents, and the AI Agent Builder still answers, using credits as it does.
• Incoming calls to your Centrus number are not answered once less than 1 credit is left, and the caller hears "This number is not available." Outbound calls fail with "This organisation has no automation credits left."
• Phone calls also need message credits. If the organisation's message credits run out, incoming calls are refused in the same way, and outbound calls fail with "This organisation has no voice credits left."
• Browser minutes run out separately. Browser Use steps then fail until the minutes reset, and the step's panel reads "New browser sessions are paused until the monthly allowance resets."

To get going again before the reset, an admin can buy more credits. They count as soon as the payment goes through.

For one agent, click Runs in the builder's toolbar. The top of the window shows how many times the agent has run and the credits it has used in all, and each run in the list shows its own credits. Select a run to see its total beside the number of steps. The Agents list shows the same lifetime total under each agent, such as 27 runs · 1,336.6 credits. Read the Runs panel explains the rest of the window.

The Agent Builder Usage report breaks the billing period down by agent, so you can see which agents use the most and why. Only admins can open it.

1. 1Click Settings in the sidebar, then the Reports tab.
2. 2Click View Report on Agent Builder Usage, the first card.

• Builder Chat is what the AI Agent Builder used while someone built or changed the agent.
• Runs is everything the agent's runs used, step by step.
• Dashboard appears when an agent has credits from outside a run. An agent's phone calls are counted here, listed as Voice call.
• Unassigned collects credits that belong to no agent, such as AI Agent Builder chats that were never saved as an agent.

Click + beside an agent to split its total into Builder chat and Runs, then click + again to list each chat turn or each run.

On a phone the report works the same way: the four totals stack and the table scrolls sideways.

09:41

Centrus

Back to Reports

Agent Builder Usage

Credit usage by agent, including Builder chat and runs, for a chosen date range.

Export PDF

Current billing periodLast 7 daysLast 30 days

Total Credits

7,846

Agents with Usage

6

Top Consuming Agent

3,450.4New Fixflo issue to BigChange job

Avg Credits/Agent

1,308

AgentStatusBuilder ChatRunsDashboardTotalLast Activity

+New Fixflo issue to BigChange jobpublished38.43,41203,450.409 Oct 2026

+Daily job summarypublished21.71,90601,927.709 Oct 2026

+Chase overdue invoicespublished52.91,336.601,389.509 Oct 2026

Swipe the table for every column

The cheapest credit is the one an agent does not need. Most savings come from making the expensive steps run less often.

• Filter early. A Filter or Condition costs 0.1, so put one before an AI step or a text, and the costly steps only run for the items that need them.
• Fetch once, not item by item. One integration step that lists 200 jobs costs 1 credit. Reading the same jobs one at a time inside a loop costs 200.
• Test with a dry run. Run now offers Dry run (no actions performed), which costs 1 credit for the whole run: the AI and the reads run for real, and anything that would send or write is only simulated. Run for real charges every step. See Test an agent before it goes live.
• Choose the model on purpose. Sorting, tagging and short replies rarely need more than GPT-5.6 Luna or Terra. Save GPT-5.6 Sol and GPT-6 Astra for the steps that need them.
• Run only as often as you need. A schedule every 5 minutes is 288 runs a day, which is 288 credits before a single step.
• Summarise rather than text everyone. An email costs 1 credit; a text costs 6 for each segment.
• Set a budget close to normal. A per-run budget a little above what the agent usually costs stops a mistake early. See Set a per-run budget.
• Check the report after a week. Open Agent Builder Usage once a new agent has run for a few days, and compare what it costs with what you expected.