> ## Documentation Index
> Fetch the complete documentation index at: https://plasma-ai.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Who pays and how limits work

The organization that owns the agent pays for its platform work. Costs are attributed to that organization, workspace, and agent—even when another member sends the request.

Organization credits are pooled. An agent's allowance is not a separate wallet funded by the person currently chatting with it.

## Understand the controls

| Control | Scope | Meaning |
| - | - | - |
| Credit balance | Organization | Available platform credits for billable usage. |
| Monthly spending cap | Organization | Limits admission of new paid work across the organization based on recorded monthly spending. |
| Monthly workspace cap | One workspace | Limits that workspace without applying the same cap to sibling workspaces. |
| Finite agent allocation | Agent and its Worker subtree | A spending allowance deducted as usage is recorded. It is not a reservation of credits or a new charge. |

All applicable controls apply together. Giving an agent a large allocation does not override an exhausted organization balance or a reached workspace cap.

An unset workspace cap adds no workspace-specific limit. A zero workspace cap pauses paid work there.

A finite agent allocation is not the same as the agent's monthly usage display. It is adjusted or replenished separately rather than automatically becoming a fresh monthly budget.

## Where to inspect usage

Organization admins can inspect **Settings → Usage**, including workspace breakdowns and spending controls.

Workspace owners with current membership can inspect their workspace's **Usage** view. Agent **Details** shows monthly model and compute usage and the agent's allowance controls where authorized.

Billing metadata does not grant access to the underlying workspace content.

## Monthly means a UTC calendar month

Spending reports and monthly caps use UTC calendar months. The interface can show exact boundaries in your local time.

Model usage follows its service time for monthly reporting; compute and corrections follow settlement time. Delayed reports can change the displayed totals.

Credit renewal follows a separate subscription calendar. Your monthly credits may renew on a different day from your spending-cap reset.

## Caps are not exact hard stops

Caps block new paid work after recorded spending reaches the limit. They do not reserve the maximum possible cost before each operation.

Work already admitted, concurrent requests, delayed provider prices, and retained sandbox storage can make spending exceed a cap or allocation. Missing usage reports do not mean zero cost.

For example, an agent can start work below its allowance and settle above it after the provider reports the final cost.

## What happens at a limit?

A reached limit can block model calls, new turns, and runtime startup, and stop managed compute. A workspace-specific cap leaves sibling workspaces subject to their own controls.

Raising a cap or funding an enabled agent can let eligible work resume. Funding does not enable an agent you explicitly disabled.

Stopped sandboxes preserve files and sessions, but retained disk storage can still accrue charges. Stopping foreground work is not the same as removing the sandbox.

## Workers and BYOK

Workers share their parent's allowance across the subtree. They retain usage attribution but do not receive independent budgets that bypass the parent.

Confirmed BYOK model costs can count toward spending caps and allocations even though those tokens deduct no platform credits. Compute remains billable. See [Credits, subscriptions, and BYOK](/fractal-product/billing/credits-and-byok).

## When the displayed balance is zero

Finishing work can produce usage past the available credit balance. Billing displays a \$0 headline balance with a separate note for that overrun, rather than a negative headline amount.

Usage past the balance is deducted from the next credit purchase. Check that note when reconciling a purchase with the resulting available credits.

## When organization credits run out

A credit cutoff stops new paid work for the organization's managed agents. What happens to a message depends on how it arrived.

| Work or content | During the cutoff | After credits are restored |
| - | - | - |
| Eligible channel messages | Activity remains pending instead of being discarded because the agent lacks credits. | The enabled managed agent automatically resumes processing pending activity, subject to access, channel controls, and other limits. |
| New direct operator messages | Sending is blocked and your draft is preserved. | Review and send the draft yourself; funding does not automatically submit it. |
| Direct requests accepted before the cutoff | Accepted request records are retained. | Retained work can proceed when eligible; do not duplicate a request simply because its response was interrupted. |
| A response already underway | The cutoff can interrupt it. | The saved session can resume, but exact continuation of that response is not guaranteed. |
| Saved conversations, Pages, and files | Saved material is retained; running out of credits does not delete it. | Continue with the retained session and files rather than creating a replacement agent. |

Saved conversations and published Pages remain readable under their existing access rules. Saved files remain retained; live file browsing still depends on the computer being available and permitted to start. A temporary unavailable file view is not evidence that the files were deleted.

Channel eligibility still matters: membership, mentions, thread routing, and channel controls determine which activity reaches the agent. Retaining pending activity does not promise a separate reply to every message.

## Restore work without overriding a manual pause

Restore the organization's credits through Billing. An enabled managed agent can automatically restart its retained sandbox and process eligible pending work when the cutoff clears.

A manually paused or **Disabled** agent stays paused. Select **Enable** when you want it to run again. Funding also does not bypass a workspace or organization spending cap, an exhausted agent allocation, lost resource access, or another runtime problem.

After an interrupted response, inspect the saved conversation and any resulting files, Pages, or external changes before asking the agent to continue. An interruption does not undo completed actions.

## Read the right billing alert

For an exhausted **Free** organization, the current notice says **Your organization has no free credits remaining**. Organization admins see **View plans**; other members are asked to contact an admin.

Paid organizations receive the organization-credit message instead. Admins can use **Go to billing page**; other members are directed to an organization admin. Missing or failed billing-plan data is not treated as proof that the organization is on Free.

Monthly spending caps and finite agent allocations are separate conditions with their own messages and recovery controls. A zero displayed balance alone does not establish that the Free-plan exhaustion notice should be shown.

See [Sending, steering, and stopping](/fractal-product/working-with-agents/controlling-work) for **Starting up**, delivery, and queued-message states.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.