Skip to main content
For the complete documentation index for agents and LLMs, see llms.txt.

Set Usage Limits (Quotas)Beta

Beta
This is coming — this page describes a feature that's still being rolled out and may not be available to you yet.

Set usage limits for metered resources in your organization, such as tokens or pipeline runs.

About This Task​

Pipelines use LLMs and external services that can be expensive. To control costs, you can set usage limits for metered resources in your organization, such as tokens or pipeline runs.

A quota on its own does not limit anything. A quota is made up of ranked policies, and together the quota and its policies work as a single unit: the ranked policies are what actually cap or allow usage. Because of this, the UI always creates, edits, and deletes a quota's policies together with the quota itself.

The Quota page has two panes:

  • Resource types: Register the metered resource (for example, a type of token usage or job) before you can create a quota for it. You cannot create a quota for a resource type that has not been registered yet.
  • Quotas (default pane): Create and manage quotas for your registered resource types.

Set Usage Limits​

  1. Click your profile icon and choose Settings.
  2. Go to Organization>Quota.
  3. Register the resource type you want to limit:
  4. Open Resource types and click Register Resource Type.
  5. Give the resource type a name and a key. The key must follow the same pattern the resource API expects.
  6. Create a quota for the resource type:
  7. Open Quotas and click Create Quota.
  8. Name the quota and choose the resource type it meters.
  9. Add one or more ranked policies. For each policy, set:
    • A label and a rank. Rank 1 is evaluated first.
    • A limit, or select Unlimited if you want the resource metered and recorded but never denied.
    • An allocation strategy: a shared pool split across all usage, a fixed limit, or an equal share divided among subjects.

View Quota Details​

Click a quota in the list to see its details. For each rank, you can see:

  • How much has been used so far
  • How much is left
  • Each subject's share of the limit
  • How many subjects the rank applies to
  • The current time window

If a rank doesn't have its first allocation yet, you'll see a note instead of a zero. The limit is still in place, and requests wait until the allocation is ready. They aren't rejected.

If the quota uses an equal-share policy that applies only to enrolled subjects, you'll also see an Enrollment section.

Edit or Delete a Quota​

To edit a quota, open it, make your changes, and click Save. Haystack Enterprise Platform saves all your changes together: it updates the quota's settings, removes any policies you deleted, updates the ones you changed, and adds the new ones.

Once a quota uses a rank, that rank stays reserved, even if you later remove the policy that used it. This means you can't reuse that rank or change the order of ranks.

To delete a quota, open it, click Delete, and confirm. This deletes the quota and all its policies.

Understand Quota Status​

The Quota page can show two statuses that indicate how quotas are currently enforced:

  • Recorded but not enforced: Usage against your quotas is being tracked, but nothing is refused yet. You see this while your organization's enforcement mode is still in legacy or shadow mode.
  • Degraded counters: The counter store Haystack Enterprise Platform uses to track consumption is temporarily unreachable, so live numbers may be stale.