AI Spend

Q: Is AI Spend my invoice?

No. Dashboard › AI Spend shows an estimate in US dollars, worked out for each AI call from the price listed for its model when the call ran. Calls without a price add nothing, so the estimate can differ from what a provider bills.

Q: What currency are the costs in?

Every amount on Dashboard › AI Spend is in US dollars, and there is no setting for another currency. Amounts under one cent are shown with four decimals, and $0.00 appears only when the cost is exactly zero.

Q: Will old costs change when prices change?

No. Each call’s cost on Dashboard › AI Spend is calculated once, when the call runs, and stored with it. A later price change affects only calls made after it, so past periods always show the same figures.

For: Account owners and anyone responsible for AI costsRequires: Any signed-in user; no plan, role or feature flag, and no AI gateway setupScreenshots show sample data

The AI Spend page shows what the calls to AI models behind your assistant cost, as an estimate in US dollars, and who and what made them: by model, stage and widget, and by customer company, customer, source, workflow, autonomous agent and run. Use it to see where the money goes, to find unusually expensive customers or runs, and to check how much of an enforced budget is left. It lives at DashboardAI Spend, and everyone in your workspace can open it, whether or not you use the AI gateway.

What you can do here

Read the estimated cost of a period. Four tiles give the estimated cost in US dollars, the number of calls and the input and output tokens, each with the exact dates it covers. Read the spend totals.
Choose which calls to count. Switch between your own agents, the services StrikeUp runs around them, or everything. Choose which calls to count.
Follow spend day by day. A chart of the estimated cost per UTC day. Follow daily spend.
Break the cost down by model, stage and widget. Three tables show which models, agent stages and widgets cost most. Break the cost down · AI models, spend and budgets.
Check how much of your budget is left. When a workspace budget is enforced as a hard limit, a box shows what it has used in its current period. Check your allowance · Budgets & limits.
See who used the AI. Tables by customer company, customer, source, workflow and autonomous agent. See who used the AI.
Find the costliest runs. The 20 workflow and agent runs that cost most. Find the costliest runs.
Change the period. The same date range as on the other Dashboard pages, capped at 90 days. Change the period.

The screen at a glance

The AI Spend page of the StrikeUp dashboard for the default range "Last 7 Days". In the header, at the right, the highlighted drop-down shows "Your agents", followed by the range drop-down "Last 7 Days", the calendar button "Sep 9 – Sep 15" and "Refresh". The "Spend" card explains that the figures are an estimate based on each provider's published prices and shows four tiles, each dated "9 Sept – 15 Sept 2026": "Estimated cost" $113.22, "AI calls" 8,287, "Input tokens" 33,991,224 and "Output tokens" 1,510,646. Below is the "Daily spend" chart in US dollars and the start of the "By model" table. — AI Spend opens on Your agents for the selected range; the drop-down switches to StrikeUp services or Everything.

Region	What it shows	What you do there
Header, right	The drop-down for which calls to count, the range drop-down, the calendar button and Refresh	Choose what to count and the period, reload
Spend card	Four tiles, the Daily spend chart and the tables By model, By stage and By widget	Read the cost and where it comes from
Usage by customer and source card	The Allowance box (only with an enforced budget), five usage tables and Costliest runs	See who used the AI and what it cost

Choose which calls to count

The first drop-down in the header has no label on screen. It decides which calls to AI models every tile, chart and table on the page counts.

Open DashboardAI Spend. The drop-down shows Your agents.
Open the drop-down and pick StrikeUp services or Everything. Both cards reload; the dates under the tiles stay the same.

StrikeUp services shows the platform's own calls around your agents, such as planning and safety checks.
What this screenshot shows
Screen
AI Spend (/admin/dashboard/ai-spend)
Highlighted
combobox “StrikeUp services”, header-right, StrikeUp services selected. every card shows only platform calls
Also visible
Tiles: Estimated cost $6.20, AI calls 10,237, Input tokens 19,458,444, Output tokens 665,682
Purpose
what the StrikeUp services plane counts

Choice	Which calls it counts
Your agents (the default)	The calls your agents make to answer visitors and do their work: the support, copilot, analytics and general agents, their sub-agents and the classifier that judges how difficult a question is
StrikeUp services	The calls StrikeUp makes around your agents: planning each answer, checking messages against your guardrails, validating policies, naming chats, summarising long conversations, choosing skills, generating toolkits and their descriptions, and analytics jobs
Everything	Every call, including calls recorded before calls were split into these two groups

Guardrail checks and planning belong to StrikeUp services, not to Your agents. Under StrikeUp services, the By stage table lists them as planner, guardrails, context_summary, title_generator and similar stages.
Everything can be more than the other two added together, because older calls with no group appear only there.
The choice is not remembered. Each time you open the page, it starts on Your agents.

The "By stage" table on the AI Spend page with the plane set to "StrikeUp services". It has the columns "Stage", "Calls", "Input tokens", "Output tokens" and "Cost" and lists the platform's own stages: planner with 3,924 calls and $3.14, guardrails with 3,945 calls and $1.40, context_summary with 718 calls and $1.36, and title_generator with 1,650 calls and $0.30. The label "By stage" sits above it and the label "By widget" of the next table below it. These calls plan answers, check messages against your guardrails, summarise long conversations and name chats. — Under StrikeUp services, By stage shows planning, guardrail checks, summaries and title generation.

Read the spend totals

The Spend card opens with the product’s own summary, An estimate of what your AI usage cost, based on each provider's published prices., and four tiles:

Tile	What it counts	Example
Estimated cost	The estimated cost of every priced call in the period, in US dollars	$115.29
AI calls	Every call to an AI model, priced or not	8,287
Input tokens	Tokens sent to the models	33,991,224
Output tokens	Tokens the models produced	1,510,646

Under every tile are the exact UTC dates it covers, such as “9 Sept – 15 Sept 2026”. They show when the 90-day limit or the current moment shortened your range.

How the cost is estimated:

Each call is priced once, when it runs, from the price listed for that model at that moment. Where a connection, such as a model aggregator, has its own price for a model, that price is used.
The cost is stored with the call. When prices change later, earlier figures stay as they were.
Some calls have no price. That happens when the provider reported no token usage, for example for a call that failed, or when no price was listed for the model at the time. These calls count in AI calls and in the tokens, but add nothing to Estimated cost, so the total is a minimum. A blue notice tells you how many calls in the period have no price; it appears in both cards.
Amounts are in US dollars only. Amounts under $0.01 show four decimals; $0.00 means exactly nothing.
Sample data that StrikeUp may add to your workspace for demonstrations is never counted here.

Follow daily spend

The Daily spend chart shows the estimated cost per UTC day across the period, in US dollars.

The "Spend" card on the AI Spend page for "Your agents". Under the subtitle about provider prices are the four tiles "Estimated cost" $113.22, "AI calls" 8,287, "Input tokens" 33,991,224 and "Output tokens" 1,510,646. The "Daily spend" area chart below shows the estimated cost per day in US dollars from Sep 9 to Sep 15, highest around $23 on Sep 9 and Sep 14. The pointer rests on Sep 12, a dashed line marks that day, and a tooltip reads "Sep 12" and "cost: 10.154563", the unrounded amount in US dollars without a currency sign. Days are calendar days in UTC. — Hover the Daily spend chart to read the estimated cost of a single day.

Hover the chart. A dashed line marks the day under the pointer.
Read the tooltip: the day, such as “Sep 12”, and its cost, such as cost: 10.3350233. The tooltip shows the unrounded amount in US dollars, without a currency sign.

Days without any call show as zero. Day labels are UTC days, so they match the dates under the tiles wherever you are. With no calls at all in the period, the chart area shows No AI usage recorded in this period..

Break the cost down

Three tables in the Spend card split the same total. Each has the columns Calls, Input tokens, Output tokens and Cost, sorted by cost, highest first.

The three breakdown tables in the "Spend" card for "Your agents", each with the columns "Calls", "Input tokens", "Output tokens" and "Cost". "By model" lists claude-sonnet-4-5 (5,836 calls, $104.38), gpt-5.1 (1,151, $7.59) and claude-haiku-4-5 (1,300, $1.25), the models the AI Gateway uses for your stages. "By stage" lists the agent stages support ($92.02), analytics ($12.35), general ($7.59) and difficulty_classifier ($1.25). "By widget" lists Demo Ltd Support chat ($88.31), Demo Ltd Help Centre widget ($21.51) and "Not attributed" ($3.40) for calls without a widget. Rows are sorted by cost, highest first. — The Spend card breaks the cost down by model, by stage and by widget; Not attributed is real spend without a widget.

Table	One row per	Things to know
By model	Model and provider	The row shows the model name only, so a model used through two providers or connections appears twice with the same name.
By stage	Agent stage that made the call	Stages show their internal names, such as `support`, `general`, `analytics` or `difficulty_classifier`, in every language. Calls without a stage are grouped as Not attributed.
By widget	Widget the call came from	A widget deleted since shows its identifier instead of its name. Calls from outside a widget, such as workflows or features of the admin app, are grouped as Not attributed.

A row with calls that have no price carries a note such as “12 without a price”. A table with nothing to show reads Nothing recorded in this period..

Example: in the screenshot, claude-sonnet-4-5 accounts for $92.02 of $115.29, the support stage for the same $92.02, and the widget “Demo Ltd Support chat” for $89.93. Most of the cost comes from one model answering visitors in one widget.

Check your allowance

The Allowance box at the top of the Usage by customer and source card appears only when your workspace has an active budget that applies to All AI usage with Enforce as a hard limit switched on. Without such a budget, the box is not shown at all.

The "Usage by customer and source" card on the AI Spend page. Under its subtitle, the "Allowance" box with a green edge reads "Workspace monthly AI budget: $250.01 of $480.00 used (52.1%)" above a progress bar whose fill is not visible, and "Resets on 1 Oct 2026"; at its top right the highlighted badge "Hard limit" shows that this budget is enforced. Below, the "By customer company" table lists Example Studios ($21.51), Generic Retail ($16.98), Beispiel Logistik GmbH ($15.85) and further companies, with "Not attributed" ($7.93) last. The allowance always shows the budget's current period, whatever range is selected. — The Allowance box appears only when a workspace budget is enforced as a hard limit; it shows the current budget period.

What it shows: the budget’s name, what it has used of its amount and the percentage, such as “Workspace monthly AI budget: $254.53 of $490.00 used (51.9%)”, a progress bar, and the date it resets, such as “Resets on 1 Oct 2026”.
The badge: Hard limit while money is left, Used up once the amount is spent.
The colour: green below 80% used, yellow from 80%, red when nothing is left.
It ignores your date range and the drop-down. It always shows the budget’s current period and counts only your agents’ calls.
Resets: a daily budget resets at midnight UTC, a weekly one on Monday at midnight UTC, a monthly one on the 1st at midnight UTC. The date is shown in your time zone, so west of UTC it can read one day earlier.
Only workspace-wide budgets count here. Budgets for one widget or one customer company do not appear in this box. If several workspace budgets are enforced, the box shows the one with the least money left.
The figures can lag by up to a minute behind the latest calls.

What happens when an enforced budget is used up, and how to set one, is explained under Budgets & limits.

See who used the AI

Five tables in the Usage by customer and source card split the cost by who or what made the calls. They have the same columns as the breakdown tables, are sorted by cost, and always list Not attributed last. They follow the drop-down for which calls to count, although the card’s subtitle ends with “Your own agents’ usage only.”

Part of the "Usage by customer and source" card on the AI Spend page, each table with the columns "Calls", "Input tokens", "Output tokens" and "Cost". The end of "By customer" shows Marcus Oyelaran ($1.92) and "Not attributed" (3,514 calls, $48.00). "By source" lists Chat (7,765 calls, $106.09), Autonomous agent (331, $4.53), Workflow (174, $2.38) and "Portal (run now)" (17, $0.23). "By workflow" lists the customer workflows Failed payments morning check ($0.91), Weekly credit note check ($0.68), Daily usage against plan limit ($0.57) and Wöchentlicher Nutzungsbericht ($0.34), and "By autonomous agent" starts with Overdue invoice follow-up (282 calls, $3.85). — These tables show who spent the tokens and how, by customer, source, workflow and autonomous agent.

Table	One row per	Things to know
By customer company	Company in your CRM	Shown by name, else by identifier
By customer	Customer	Shown by name, else email, else identity; anonymous visitors appear with an identity beginning with `anon-`
By source	Where the call started	Chat, Workflow, Autonomous agent or Portal (run now)
By workflow	Workflow	Scheduled runs and runs started from the customer portal; a deleted workflow shows its identifier
By autonomous agent	One of your autonomous agents	A deleted agent shows its identifier

Not attributed is real spend that could not be tied to a customer, source, workflow or agent: calls made before this information was recorded, calls from features of the admin app, and anonymous chats without a customer. It is included in every total.

Find the costliest runs

The Costliest runs table lists up to 20 workflow and autonomous-agent runs in the period, most expensive first.

The "Costliest runs" table at the bottom of the AI Spend page, with the columns "Run", "Source", "Customer", "Status", "Calls", "Tokens" and "Cost". Each row shows the first characters of a run id and its date, for example 13fd0425 from 14 Sept 2026 with 46 calls, 272,780 tokens and $0.61. The Source column reads "Not attributed" in every row, with the workflow or autonomous agent name below it where there is one, such as Weekly credit note check, Overdue invoice follow-up or Failed payments morning check; chat runs have no name. The Customer column names people such as Rosie Kendall and Morgan Hayes with their company below, and Status shows "Completed" in grey or "Failed" in red. — Costliest runs lists up to 20 workflow and agent runs by cost; read the name under Source to see what ran.

Column	What it shows
Run	The first 8 characters of the run’s identifier and the date the run was created
Source	Not attributed in every row, with the name of the workflow or agent below it
Customer	The customer the run worked for, with their company below; Not attributed when there is none
Status	The run’s status in English: `Planned`, `Running`, `WaitingForApproval`, `WaitingForInput`, `Paused` (yellow), `Completed`, `Failed` (red) or `Cancelled`
Calls	Calls to AI models the run made
Tokens	Input and output tokens together
Cost	The run’s estimated cost

Read the name under the Source column to see what ran. To follow a run step by step, find it by its identifier in AnalyticsAgent jobs.

Change the period

The range drop-down and the calendar button set the period for both cards, except the Allowance box. The range is shared with the other Dashboard pages and remembered in this browser. On this page two extra rules apply:

At most 90 days. A longer range is cut to the 90 days before its end. A range from 1 March to 30 September covers 2 July to 30 September.
It ends now. A range that ends today stops at the current moment.

The dates under the tiles always show the period actually counted.

Every control, explained

Drop-down for which calls to count

Where

The first drop-down at the top right, with no label on screen. It applies the moment you pick an option.

What

Your agents (the default), StrikeUp services or Everything. It applies to both cards, but not to the Allowance box. It resets to Your agents on every visit.

Why

Your own agents’ calls are the ones your setup controls. The other choices show the full picture of AI use around your assistant.

Range drop-down and calendar

Where

To the right of the first drop-down. They apply the moment you choose.

What

Last 7 Days (the default), Last 14 Days, Last 30 Days or a custom range from the calendar, up to today. On this page the range is cut to 90 days and ends at the current moment. The choice is shared with the other Dashboard pages and stored in this browser.

Why

Match the period to the one you are checking, such as a calendar month.

Refresh

Where

The Refresh button at the top right.

What

Reloads both cards for the same choice and range; the button spins until both have loaded.

Why

See the cost of calls made since you opened the page.

Spend card

Part	What it shows	Why
Estimated cost, AI calls, Input tokens, Output tokens	Totals for the period, with the UTC dates counted under each	The headline cost and volume
Notice about calls without a price	How many calls in the period have no cost estimate	Tells you the total is a minimum
Daily spend	The estimated cost per UTC day, axis USD	Spot expensive days
By model, By stage, By widget	Cost split by model, agent stage and widget	Find where the cost comes from

Usage by customer and source card

Part	What it shows	Why
Allowance	The enforced workspace budget with the least money left: used amount, percentage, badge and reset date	Know whether work will pause soon
By customer company, By customer	Cost per customer company and per customer	Find the customers who use the most AI
By source	Cost per source: chat, workflow, autonomous agent or portal run	See which kind of work costs most
By workflow, By autonomous agent	Cost per workflow and per autonomous agent	Find automations to tune
Costliest runs	Up to 20 runs by cost	Find single runs that cost far more than usual

What you need to know

Topic	Fact
An estimate	Costs are worked out from listed model prices at the time of each call. They are not an invoice.
Currency	US dollars only.
Frozen prices	A call’s cost is stored when it runs; later price changes do not alter it.
A minimum	Calls without a price add nothing to the cost total.
90 days	The period is at most 90 days and ends at the current moment. Other Dashboard pages have no such limit.
UTC days	Days and the dates under the tiles are UTC.
No sample data	Demonstration records are never counted on this page.
Who sees it	Every signed-in user of your workspace, whatever their role, including customer names and costs.
No export	The page has no download and rows are not clickable.
Live data	Every load, change and refresh reads current data. The allowance box can lag by up to a minute.

Frequently asked questions

Cost

Is AI Spend my invoice?

No. DashboardAI Spend shows an estimate in US dollars, worked out for each AI call from the price listed for its model when the call ran. Calls without a price add nothing, so the estimate can differ from what a provider bills.

What currency are the costs in?

Every amount on DashboardAI Spend is in US dollars, and there is no setting for another currency. Amounts under one cent are shown with four decimals, and $0.00 appears only when the cost is exactly zero.

Why do some calls have no price?

A call has no price when the provider reported no token usage, for example because the call failed, or when no price was listed for that model at the time. Such calls count in AI calls and the token tiles but add nothing to Estimated cost. Your AI budgets still charge them, at a typical rate, so a budget can show more spend than this page. Calls through your own Self-hosted endpoint connector to a model without a price are not unpriced: they are recorded at $0.

Will old costs change when prices change?

No. Each call’s cost on DashboardAI Spend is calculated once, when the call runs, and stored with it. A later price change affects only calls made after it, so past periods always show the same figures.

Why does the same model appear twice in By model?

The By model table has one row per model and provider, but shows only the model name. When the same model is reached through two providers or connections, for example directly and through an aggregator, it appears in two rows with the same name.

What is counted

Why is Everything not the sum of Your agents and StrikeUp services?

Everything counts every AI call, including older calls recorded before calls were split into Your agents and StrikeUp services. Those older calls belong to neither group, so they appear only under Everything.

Are guardrail checks part of Your agents?

No. Checking visitor messages against your guardrails, planning answers, naming chats and summarising conversations are counted under StrikeUp services on DashboardAI Spend. Pick that option in the first drop-down to see their cost by stage.

What is a stage?

A stage is the agent or step of StrikeUp that made the AI call, such as support, general, analytics or, under StrikeUp services, planner and guardrails. The By stage table shows these internal names in every language.

What does Not attributed mean?

Not attributed groups real spend that could not be tied to a stage, widget, customer, source, workflow or agent, such as calls made before that was recorded, calls from admin-app features and anonymous chats. It is included in every total and listed last in the usage tables.

The By widget table shows a widget’s name while the widget exists. When the widget has been deleted since the calls were made, its cost stays in the table under the widget’s identifier instead of its name.

Does AI Spend include sample data?

No. Sample records that StrikeUp may add to a workspace for demonstrations are never counted on DashboardAI Spend, although the other Dashboard pages do count them.

Why does Costliest runs show Not attributed in every Source cell?

The Source column of Costliest runs shows Not attributed for every run. The name of the workflow or autonomous agent that ran is shown on the line below it, and By source splits cost by source correctly.

Period

Why is my range shorter than I picked?

DashboardAI Spend counts at most 90 days, the 90 days before the end of your range, and a range that ends today stops at the current moment. The dates under each tile show the period actually counted.

Why do some days in Daily spend show zero?

The Daily spend chart has one point per UTC day of the period. A day with no calls to AI models at all, or only calls without a price, is shown as zero rather than left out.

Why does the drop-down go back to Your agents?

The first drop-down on DashboardAI Spend is not remembered between visits. Every time you open the page it starts on Your agents, while the date range keeps your last choice.

Allowance

Why don’t I see the Allowance box?

The Allowance box appears only when your workspace has an active budget that applies to All AI usage with Enforce as a hard limit switched on. Budgets that only alert, or that apply to one widget or one customer company, do not show it.

Does the Allowance box follow my date range?

No. The Allowance box always shows the budget’s current period, daily, weekly or monthly, counting only your agents’ calls, whatever range or drop-down choice is set on DashboardAI Spend.

When does the allowance reset?

At the start of the budget’s next period in UTC: the next midnight for a daily budget, Monday at midnight for a weekly one, and the 1st of the month at midnight for a monthly one. The Allowance box shows the date in your time zone.

Access

Do I need the AI gateway to see AI Spend?

No. DashboardAI Spend is there for every workspace, with or without the AI gateway, because every AI call is costed. Setting up your own models and keys is described in AI models, spend and budgets.

Who can see AI Spend?

Every signed-in user of your workspace can open DashboardAI Spend, whatever their role, and sees the costs together with the names of customers and customer companies. End customers never see the Dashboard.

Troubleshooting

Symptom	What to check
A Request failed message and No AI usage recorded in this period.	The page failed to load; it is not an empty period. Select Refresh, and sign out and in again if it happens again.
Estimated cost is $0.00 although AI calls is above 0	Every call in the period has no price; check the blue notice. Under StrikeUp services, models without a listed price produce the same result.
The estimate is lower than your provider’s bill	Calls without a price add nothing; the first drop-down may be on Your agents only; and providers can round or charge minimums differently.
The Allowance box is missing	No active workspace-wide budget has Enforce as a hard limit on. Check under ConfigureBudgets & limits.
The Source column of Costliest runs reads Not attributed	Every row shows this. Read the workflow or agent name below it, or use By source.
The period shows fewer days than you picked	The page counts at most 90 days and stops at the current moment. Read the dates under the tiles.
A large share is Not attributed	Calls from admin-app features, anonymous chats and older calls have no customer or source. Identify your users in the widget so their chats are tied to a customer.
A widget appears as a long identifier	The widget was deleted after the calls were made.

AI Spend

What you can do here

The screen at a glance

Choose which calls to count

Read the spend totals

Follow daily spend

Break the cost down

Check your allowance

See who used the AI

Find the costliest runs

Change the period

Every control, explained

Drop-down for which calls to count

Range drop-down and calendar

Refresh

Spend card

Usage by customer and source card

What you need to know

Frequently asked questions

Cost

Is AI Spend my invoice?

What currency are the costs in?

Why do some calls have no price?

Will old costs change when prices change?

Why does the same model appear twice in By model?

What is counted

Why is Everything not the sum of Your agents and StrikeUp services?

Are guardrail checks part of Your agents?

What is a stage?

What does Not attributed mean?

Why is a widget shown as a long identifier?

Does AI Spend include sample data?

Why does Costliest runs show Not attributed in every Source cell?

Period

Why is my range shorter than I picked?

Why do some days in Daily spend show zero?

Why does the drop-down go back to Your agents?

Allowance

Why don’t I see the Allowance box?

Does the Allowance box follow my date range?

When does the allowance reset?

Access

Do I need the AI gateway to see AI Spend?

Who can see AI Spend?

Troubleshooting

Related