AI credits
AI Credits are the prepaid balance a workspace draws from when people use Rescale's AI features. Two pages in the administration portal report on that balance: a dashboard showing how credits are being consumed, and a financials page listing the individual credits the account holds.
Two products draw on AI Credits. Agentic Engineering covers Rescale Edison and agents, and AI Physics covers model training. Each product has its own pool, everyone in the workspace draws from the shared pool for that product, and the two cannot be spent interchangeably. They are also metered differently, which is why their consumption patterns look so different: Agentic Engineering is drawn down by the size of each interaction, while AI Physics is drawn down at fixed hourly rates.
Agentic Engineering runs entirely on AI Credits, so every Edison and agent interaction draws from the pool and the dashboard is the only place to watch it.
Advanced Modeling & Simulation includes an allowance of Edison usage each day instead, so a workspace on that tier can operate without a credit balance at all. It can also buy AI Credits on top of the daily allowance.
AI Physics credits are a separate pool, bought separately and independent of the other. A workspace can hold one pool, both, or neither, which is why these pages can look empty on an otherwise healthy account.
If you already manage Rescale deposits, AI Credits will feel familiar. They are a prepaid balance with threshold notifications and consumption reports, scoped to AI features rather than to compute.
Before you start
You need to be an organization admin or a workspace admin, and access follows the admin role you already hold.
The workspace also has to be licensed for Agentic Engineering, AI Physics, or both. A workspace on Advanced Modeling & Simulation does not use a shared AI Credit pool and will not see either of these pages at all.
Dashboards > AI Credits Utilization
Open the organization administration portal, or the workspace portal if you only want one workspace, and select Dashboards > AI Credits Utilization in the left navigation. The month picker in the title bar sets the period, and defaults to the current month.
The page is made up of four sections.
Balance cards
You get one card per product pool the workspace is licensed for, so an account with both shows Agentic Engineering and AI Physics side by side. Each card reports the total credits provisioned to that pool, including both the license allocation and any top-ups, what remains to spend, and how much has been consumed to date. A bar and a percentage beneath the figures give you the same information at a glance.
Read the two cards independently, because the pools are tracked and spent separately. A healthy Agentic Engineering balance tells you nothing about whether AI Physics training is about to be blocked.
Daily Burn Chart
A column chart of credits consumed per day across the selected month, with a separate colored series for each licensed product. Days with no consumption are left empty.
The two series have different shapes because the products meter differently. Agentic Engineering rises and falls with how much people are asking and how much context each interaction pulls in. AI Physics tracks hours of training work, so one long training run can produce a tall column on an otherwise quiet day. Comparing the two is how you tell which product drove a heavy month.
Consumption by User
A table of the users who consumed credits during the selected month, ranked from highest to lowest, five rows to a page. It reports each user's email, the credits they drew from the Agentic Engineering pool, and the credits they drew from the AI Physics pool. At organization level a Workspace column identifies where the consumption came from.
Selecting View Billing Summary opens the billing summary for the same scope.
Prior Periods
Consumption totals for the three months before the one you are viewing, so you can see a trend without moving the month picker repeatedly.
Financials > AI Credits
A member without an admin role does not see the workspace pool at all. They see their own consumption and quota state through Edison instead, so "how much is left" is an administrator's question and "how much have I used" is anybody's.
Where the dashboard shows how credits are being spent, Financials > AI Credits lists the credits themselves. Each row is one independent purchase, whether that is a license allocation or a top-up bought later, and each carries its own consumption report.
A workspace typically holds one credit from its license allocation plus a further row for each top-up purchased since.
Exporting consumption data
There are two exports and they answer different questions.
The export link on the Consumption by User card pulls per-user consumption for the month you are viewing. That is the one to use when you want to know who is spending.
The CSV link on a row under Financials > AI Credits gives you a consumption report scoped to that single credit, with columns for month, workspace ID, user, type, amount and description. That is the one to use when you are reconciling a specific purchase.
How current the figures are
Consumption is rolled up on a schedule several times a day rather than continuously, and that affects the parts of these pages differently.
The balance figures account for the current day's usage as well as completed rollups, so Remaining is close to real time. The burn chart, the Consumption by User table, Prior Periods and both CSV exports are all built from completed rollups only, so activity from the last few hours may not appear in them yet.
If you are checking whether a particular Edison conversation, agent run or AI Physics training job was counted, look again the following day rather than concluding it was missed.
Threshold notifications
Admins are emailed automatically as a pool draws down.
These use the same recipient list and preferences as deposit threshold notifications, which are configured at Preferences > Notifications and described under spend notifications. They are limited to one email per purchase of credits per day. Each pool notifies on its own thresholds independently, so a workspace can sit at 80% on Agentic Engineering and 20% on AI Physics while admins only hear about the first.
When a pool reaches zero, the features it funds stop for every user in the workspace until more credits are added.
If Agentic Engineering is depleted, Edison and agent interactions stop. AI Physics work carries on provided that pool still has credits.
If AI Physics is depleted, any running workstations doing AI Physics training and any running jobs doing AI Physics batch work are stopped, and new ones are blocked from starting. Edison and agents keep working provided that pool still has credits.
In both cases the rest of Rescale is unaffected. Submitting and running standard jobs, storage and software all continue as normal.
Related
- Spend notifications — the same threshold mechanism for budgets and deposits
- Billing & reports — reconciling AI charges against the rest of your spend
- Rescale Edison — what Agentic Engineering credits are spent on
- AI Physics — what AI Physics credits are spent on