This article is auto-synced from its in-app version in Tai.
Company admins only. The Billing Center is visible to users with the Company Admin role. The per-chat and per-report Credit figures described further down are visible to everyone.
Usage Credits (short "Credits") are the unit on which the Usage Fees of TAI are based. The Billing Center brings your company's Credit consumption together in one place — what you consumed, what it was spent on, and which work drew the most.
The Billing Center is organized as three sidebar entries: Overview for the breakdowns and charts below, Top sessions for the per-session ranking, and Budgets, where you can cap how many Credits your company spends over a day, week or month and get warned — or optionally stop new usage — before you go over. See Credit budgets for details on setting one up.
Note: The descriptions on this page are provided for general documentation purposes and may differ slightly from formal definitions. For exact and legally binding Credit and billing terms, please refer to the Customer Terms of Service for Tai.
When you'd use this
At the end of a month, to see how many Credits your company consumed and where they were used. Or mid-month, you can check whether a new agent is consuming more Credits than you expected. For a single piece of work — one long chat, one weekly report — the Credit figure sits on the chat or report itself.
Credit Consumption
Credit consumption comes from two situations:
Session Initiation: Each session you start consumes a certain number of Credits based on the agent you use.
Qualified Actions: The following types of Qualified Actions exist:
Session Extensions, which are events where an agent session has exceeded a defined threshold of AI compute and starts a new Unit of Work.
Connector Calls.
Specifically defined actions of built-in agents, which are defined in the Billing Center together with their Credit consumption.
Billable and Included
Credits are divided into Billable Credits — the part we charge for — and Included Credits, which you consumed but aren't charged for. Total value is the two added together. Included Credits are shown so you can see the full value of what you used, not just the part that counts against your plan.
Billable is not the same as invoiced. It is what your usage costs under your plan, not an amount owed — Invoiceable Credits below explains the difference.
Session Extensions: Units of Work
AI compute measures the work an agent does. Once the defined threshold of AI compute (called a "Unit of Work") has been exceeded — which can happen with a long conversation, many follow-up questions, or a lot of data to reason through — the session needs to be extended with a new Unit of Work.
The category of the model the agent runs on defines how many Credits a Unit of Work (i.e. a session initiation or session extension) consumes. A more capable model costs more per Unit of Work:
Fast — 1 Credit per Unit of Work
Balanced — 3 Credits per Unit of Work
Most capable — 5 Credits per Unit of Work
Other — 3 Credits per Unit of Work. A model we haven't placed in one of the three categories is charged at the Balanced rate.
In the Billing Center, the By capability breakdown groups a session's whole Credit figure under the category the agent ran on, so its rows are larger than the AI compute row in By category.
Connector Calls
Every call an agent makes to a connector is priced by that connector's billing tier. The tier is shown on the connector itself, and the Billing Center's By connector tier breakdown groups your spend by it.
Included — free. Today this covers our own TripBuilder APIs: TripBuilder Cockpit API and TripBuilder Cockpit Areas API.
Default — 1 Credit per call. Everything not marked Included or Premium, including connectors you set up yourself.
Premium — 2 Credits per call. External data feeds we pay for, such as OAG Aviation.
The tier prices the call TAI makes; it says nothing about what the data costs you elsewhere. A connection you pay nothing for on the provider's side — one where Nezasa supplies the access and there is no key for you to buy — still consumes connector Credits at its tier for every call an agent makes to it.
Included calls are where most of your Included Credits come from. Such a call still shows a Credit value — 1 Credit, the same as a Default call — but it isn't charged, so you can see both what the call was worth and that you got it for free. The free storage allowance below works the same way.
A Default or Premium call can land in Included too. When TAI can answer it entirely from its own response cache — no fresh request to the connected system — the call costs nothing extra to serve, so it is shown as Included instead of Billable, at its tier's usual Credit value. Only a fully cached call qualifies: one that still had to fetch part of its answer live is billed as usual. For example, a Default connector called 20 times in a session, 6 of them served entirely from cache, shows 14 Billable and 6 Included Credits for that connector — 20 Credits of value either way. This applies to sessions from 1 September 2026 onwards.
Other Qualified Actions
Built-in agents can have specific Qualified Actions defined which consume a certain number of Credits. Such Qualified Actions, together with their Credit consumption, are listed in the Billing Center.
Storage Credits
Storage is a standing charge, not a per-use one. What you hold is measured once a day and charged for that day, so a month is the sum of its days rather than an average — holding 10 GB for a week costs the same whether or not anyone searched it.
The rate is 1 Credit per whole GB per day, and the first GB is free. Storage is rounded up to whole gigabytes (1 GB = 1,000,000,000 bytes), so a company holding 9.2 GB is counted at 10 GB: 1 GB included, 9 GB billable, 9 Credits a day.
What counts is the knowledge you own — the text of your knowledge sources, the embeddings that make them searchable, and the source files you uploaded. Nezasa's built-in knowledge doesn't count, and neither does knowledge you have deleted.
The rates
What | Detail | Credits |
AI compute, per unit of work | Fast model | 1 |
AI compute, per unit of work | Balanced model | 3 |
AI compute, per unit of work | Most capable model | 5 |
AI compute, per unit of work | Other model | 3 |
Connector call | Included connector | 1, not charged |
Connector call | Default connector | 1 |
Connector call | Premium connector | 2 |
Connector call | Any tier, served entirely from cache | Its tier's value, not charged |
Storage | Per whole GB held, per day — first GB free | 1 |
Days, months and long conversations
Credits are counted on the day the work happened. Each day is confirmed once it has ended, and the most recent confirmed day is re-checked once more, so that work which arrived just before midnight lands on the day it belongs to. After that a day stays as it is. A month is the sum of its confirmed days, and usage is settled a calendar month at a time.
Conversations never have to end for this to work — TAI does not wait for a chat to be finished. A chat you leave open and come back to a week later is counted on each of the days you actually used it. So a conversation that starts at the end of one month and continues into the next is counted separately in each month: what it consumed in the first month belongs to that month, what it consumed in the second belongs to the second. A conversation that keeps going is split across as many months as it touches.
Splitting a conversation this way never costs more. Units of Work and connector calls are counted across the whole conversation exactly as they always are — only which month they land in follows the calendar, so every month's figure is complete on its own and no consumption falls between two of them. A single-day chat is simply the case where all of that lands on one day.
Budgets watch days; usage is settled by month. A credit budget and your monthly figures are two views of the same consumption at different grains. A budget tracks Credits day by day over a daily, weekly or monthly period so you can see a ceiling approaching and act before you reach it. Settlement takes the same Credits a calendar month at a time. The two never disagree about what a day cost — they differ in the window they add it up over, so a budget period that is not a calendar month, or a day that is still being confirmed, is enough to explain a figure that looks different.
Invoiceable Credits
The Billing Center reports what you consumed. What you are invoiced for is a different question. The page answers one part of it: when the Credits on screen aren't invoiced — or only partly are, because your date range crosses the day your billing started — a notice at the top says so, and why. What an invoiced period actually costs you, it doesn't answer. Two things sit between the two.
Total value is Billable plus Included. Included Credits are never invoiced — they are shown so you can see the full value of what you used. Only the Billable figure can reach an invoice at all.
Your agreement may include a monthly credit allowance. When it does, the Billable figure on the Overview carries a small allowance note in its top-right corner (see below). Credits inside the allowance still count as Billable in the breakdowns, because those report consumption rather than your contract — the note is where you see how much of the allowance this month's usage has reached.
So a Billable figure is what your usage costs at the standard rates — not an amount owed. What you actually pay depends on your agreement: the allowance in it and the rates your company has agreed, which may differ from the ones above.
For anything commercial — your rates, your allowance, or a figure you would like explained — talk to your Nezasa account manager.
Your monthly credit allowance
If your TripBuilder contract includes a monthly credit allowance (for example 500, 1,500 or 5,000 Credits a month), the Billable figure at the top of Overview shows a small note in its top-right corner: Allowance this month, with this month's Billable Credits so far against the allowance — for example 249 of 1,500 credits used. Hover over it (or tab to it) for the details:
Resets — the day the allowance starts over (1st of the month, UTC).
Upcoming change — if a change is scheduled (a new amount, an end, or an allowance that hasn't started yet), the tooltip says from which month and to what. While an allowance has not started yet, the note itself reads Monthly allowance and shows when it starts, for example
From Dec 2026: 5,000.
How it works:
The allowance is per calendar month (UTC). It absorbs the earliest Billable credits of the month first. Unused credits expire at month end and never carry over.
Included usage — free connectors, the first GB of storage — is never counted against it. Allowance credits are their own pool next to Billable and Included.
The note covers this month only, whatever date range you picked below, and it is provisional until the month closes: it counts today's usage as it happens.
The note is hidden when your company has neither a current nor an upcoming allowance, or when this month isn't being invoiced for your company (see the notice at the top of the page).
On Budgets, the monthly budget shows the allowance next to its limit. Budgets themselves are unchanged: they still count every Billable credit, allowance or not.
Steps
In the Workbench sidebar, click Billing Center.
Pick a date range. The page opens on the last 30 days. The range is set per view — switching between Overview and Top sessions starts each from the last 30 days again.
On Overview, check the allowance note in the Billable tile if you have one, read the three headline figures, then the breakdowns beneath them — by category, capability, identity, connector tier, agent ownership and agent. Each shows a table beside a chart.
Switch the bucket picker between Daily, Weekly and Monthly to regroup the charts. The change is instant — it regroups what is already loaded.
Narrow to one or more agents with the agent filter to see what a particular agent consumed.
Click Top sessions in the sidebar to rank individual chats, reports, agent invocations and Advisor runs by Credits, biggest first.
Click Download CSV for the rows behind the Overview — one per session per day.
What a typical session costs
The breakdowns above answer "where did the Credits go". Two tables at the bottom of Overview answer a different question: how expensive a typical session is, so you can tell whether a big number was one unusual conversation or the normal cost of the work.
Credit distribution by agent and Credit distribution by user each show, per row, the average, median, smallest and largest Credit figure across that row's sessions, plus how many sessions there were. The median beside the average is the point: if they are close, the row's sessions cost much the same and the average is a fair estimate; if the average is well above the median, a few expensive sessions are pulling it up, and the maximum tells you how far.
Both cover the date range you selected. A session that started before the range counts the Credits it added inside it.
Personal agents are left out of the by-agent table — an agent someone created for themselves isn't a company-wide cost profile. They still appear in the breakdowns above, and their sessions still count towards the person's row in the by-user table. The table says so beneath its title.
The by-user table covers people. Work an API key did has no person to attribute it to and isn't listed, so its session count can be lower than the totals above.
The by-agent table is visible to everyone who can open the Billing Center; the by-user table requires the Company Admin role.
Where you see Credits
A chat — under the last message. It covers the whole conversation, not the last turn, and grows as the conversation does — including while the agent is still writing, so a long answer's cost is visible as it accrues rather than only at the end.
A report — in the Consumption section of the report or run detail, alongside when it ran.
A figure shown with a ≈ is still being finalised. Credits are confirmed once a day, so recent work shows an estimate first. The estimate is calculated the same way as the final figure — it moves only because the work itself is still growing.
A chat's line also shows how many connector calls the conversation made. Hover it for the detail behind both figures: the Credits split into AI compute and Connectors, and the calls split by connector, so you can see which system an agent leaned on. On a report, the same detail sits in the Consumption section, no hovering needed.
The figure on a chat or a report is a consumed total — the same Total value the Billing Center reports, Billable and Included added together. A conversation that called an Included connector carries those calls' value in its total without being charged for them; one that called none is billable end to end and the total is the same either way. Only the Billing Center splits the two.
Tips and limits
Credits settle once a day. A session run today shows a live figure marked ≈ in Top Sessions, the same convention described above, until the day is confirmed. The Overview figures don't wait either: they include today's work as it happens.
A session that starts before, or continues past, your selected date range is marked Partial in Top Sessions. Its turn and activity counts cover only the part inside the range — hover the marker to see how many of the session's events that is. If the session began before your range, hovering the date beside the marker shows its real start. Its Credit figure is unaffected — it always covers the full session.
The CSV has one row per session per day. A conversation that ran across three days is three rows, each carrying the credits that day added — so its rows added together are the conversation's total.
The CSV covers confirmed days only. It excludes days still being confirmed and storage, so its total will be lower than the Overview's. Use it for closed periods, not for the last day or two.
The CSV's Credit columns are named
billable, notbilled.billed_credits,compute_billed,connectors_billedandstorage_billedare nowbillable_credits,compute_billable,connectors_billableandstorage_billable— the same figures under the same names the page uses. If you import this file into a spreadsheet or script, point it at the new column names.The agent filter applies to the Overview and to the CSV, not to Top Sessions. The export is the Overview's rows, so it narrows with the filter — clear the filter first if you want the whole company. Top Sessions always ranks every session in the window.
Storage is left out while an agent filter is active. Storage is billed per company and has no single agent to attribute it to — it is a standing daily charge on your whole knowledge footprint, not something any one agent consumed. The page says so when the filter is on.
An empty period means nothing has been confirmed for it yet, not that the work was free.
Credit figures appear in TAI itself. A chat embedded in TripBuilder or another application of yours shows none — the consumption is counted the same way, you just read it here.
The Billing Center shows consumption. It is not an invoice, and the figures are not an amount owed — see Invoiceable Credits above.
Related

