Usage and budgets
See every AI request, what it cost and which model ran it, follow a task step by step, and set spending limits.
Pulse keeps a record of the AI work done on your behalf: in chat, in the editor, when your documents are prepared for searching, by Tron agents on the shared Pulse connection, and by AI features in other modules. You can see what each request cost and which model handled it, and set limits so AI work stops before it goes over budget.
Everything on this page lives in Settings → Usage.
Choose the period
At the top right of Usage, choose 7d, 30d or 90d to see the last 7, 30 or 90 days. Everything on the page follows that choice.
The overview
The tiles at the top show the totals for the period:
- Attempts is how many AI requests were made.
- Tokens in and Tokens out are the text sent to models and the text they wrote back.
- Cached / reasoning counts cached and reasoning tokens, as each provider defines them.
Below the tiles, Cost certainty splits cost by how sure Pulse is of it, and Requests by outcome shows how many requests completed, failed, were cancelled or were interrupted.
How sure a cost is
Each request’s cost carries one of these states:
| Cost state | What it means |
|---|---|
| Actual | The provider reported exactly what was used. |
| Estimated | Pulse had to estimate the usage. |
| Unknown price | The model has no price set up, so Pulse doesn’t report a cost for it. It never counts it as free. |
| Not billable | The request didn’t cost anything to bill. |
Pulse never adds costs together across different states or currencies, so an estimate is never passed off as an exact figure.
Find a request
Use the filters to narrow things down. You can filter by date (From and To), Connection, Model, Module, Use case, Status, Cost state and Currency. Select Clear all filters to start again.
Then pick a panel:
| Panel | What it shows |
|---|---|
| Requests | Every request, newest first, with its time, module, model, tokens, cost and status. Select Load more for older ones. |
| Breakdowns | Totals By model, By connection, By module and By use case. |
| Documents & indexing | What it cost to prepare your documents for searching, by stage and by folder. |
| Budgets | Your spending limits, and how much of each is used. |
| Provider errors | Requests that failed, were cancelled or were interrupted, with the error in plain words. |
See one request in detail
- On the Requests panel, select a row.
- The Request detail panel shows its status, model, tokens, cost, how long it took and which connection it used.
- Select Open task trace to see the whole task this request belonged to.
Export your usage
Select Export CSV or Export JSON to download the requests. The export uses the filters you’ve set. It contains usage and cost figures only, never the text of prompts or answers.
Follow a task step by step
A single answer can take several AI requests, for example planning, writing and checking. The task trace shows them all together.
- Each step lists its time, component, model, status and cost. How this step ran explains which model was chosen and, if a task budget is set, how much of it is left.
- The total at the top is the sum of the steps below it.
- If a Tron agent asked for approval during the task, the step shows that approval and its status, such as approved or waiting.
Select Back to requests to return to the list.
Good to know You give or refuse approvals in Tron, not in Pulse. Pulse only shows their status here. See Tron.
Spending limits
Pulse checks every request against your limits before sending it to a model. If a request would go over a limit, it isn’t sent, and you see a message saying the budget is used up. Nothing is spent on a request that’s been stopped.
A budget for each task
You can cap how much any one task may use, such as a single chat answer or a single agent run.
- Open Settings → AI connections.
- Expand Advanced settings and select Edit routing and limits…
- Under Budget for one task (optional), fill in any of Token limit, Request limit and Spending limit. For a spending limit, also set the Currency.
- Under Models with no configured price, choose what Pulse does with models it can’t price: Decide from the spending limit, Never use them, Use them, capped by tokens or Use them.
- Select Save policy.
Note A model with no configured price isn’t free. Pulse can’t measure what it costs, so a spending limit can’t protect you from it. Choose Never use them if you want a strict cap.
Workspace budgets
The Budgets panel on Usage lists each budget that applies to your workspace, with its Limit, how much is Used, how much is Reserved for requests still running, and how much is Uncertain because a provider never reported its usage. If no budget is set up, it says No budget is configured for this workspace.
Pulse holds back the expected cost of a request while it runs. If a provider never reports what a request used, that amount stays counted as uncertain rather than being handed back as if it were free.
Content logging
By default, Pulse doesn’t store the text of your prompts or answers. The usage records hold counts and costs only.
The workspace owner can turn Content logging on at the bottom of Usage:
- Set Keep for (days), from 1 to 365.
- Select Turn on, then confirm with Turn content logging on.
When it’s on, Pulse stores prompts and responses for everyone in the workspace, not only the owner. Keys and tokens are removed before anything is stored, and hidden model reasoning is never stored. Stored text is deleted at the end of the retention period.
To stop, select Turn off. To remove everything stored so far, select Delete logged content now, then Delete it now. Usage totals, traces and costs stay as they are.
Note Only the workspace owner can change content logging. Everyone else sees whether it’s on and for how long text is kept.
Tron agents and other modules
Usage from outside Pulse appears here too:
- Tron agents on the shared Pulse connection. In Tron’s Agent Builder, under Execution, set Runs on to Shared Pulse connection and choose an execution policy. Pulse then picks the model, applies your limits and records the cost. See Tron.
- AI features in other modules, such as Cue’s AI website generation. You can switch these off in Settings → Module intelligence. See Get started with Pulse.
Filter by Module to see what each one costs.
Related: Choose and connect AI models · Frequently asked questions