AI GridDocs
Sign Up

Spending and content policies

Set organization- or project-scoped spending limits, output caps and content guardrails that apply to every call.

Updated Sep 9, 2026

Open policies#

Open Policies under the API group in the left rail. The table shows:

Column Meaning
Policy Name
Scope Entire organization or one project
Spending limits Daily / monthly USD ceilings
Max output Output token cap
Guardrail Off / Observe / Mask / Block plus rule packs
Status Enabled or disabled

Row menu: Edit, Disable (or Enable), and Delete with an optional reason.

Policies table with scope, spending limits, max output, guardrail and status columns
The Policies table — scope, daily/monthly spending limits, output cap, guardrail mode with its rule packs and status; New policy is the header action.

Create a policy#

Click New policy and fill the drawer:

  1. Name — something you recognize in audit later.
  2. Scope — Entire organization or a single project. Personal workspaces hide project setup.
  3. Allowed products — an optional allowlist. When several active policies name allowlists, the lists intersect: a call must be in every list to pass.
  4. Daily limit / Monthly limit — USD ceilings on charged spend. Outstanding (still-held) reservations count toward the ceiling even across UTC day/month boundaries.
  5. Max output tokens — hard cap on reply length.
  6. Content guardrail mode + Rule packs (up to 3):
    • Off — no screening.
    • Observe — matches are counted and audited, nothing changed.
    • Mask — matched content is masked before dispatch.
    • Block — a match refuses the call before any reservation.

Click Save. The policy applies to the next call — admission re-evaluates active policies inside the transaction, so a change takes effect immediately, even on requests already in flight elsewhere.

Creating a policy — scope, spending limits, output cap and a content guardrail — and how it takes effect on the next call.

How limits combine#

  • All active applicable policies remain in force at once. A policy is a restriction, not a wallet or an authority grant: it can only narrow what callers may do, never add access.
  • A call must fit every applicable ceiling: daily/monthly spend, max output, and the intersected allowlists.
  • Guardrail packs screen user-submitted text with pattern heuristics (PII, secrets, injection). A blocked call is refused before any money moves.

Honest limits#

  • Policy ceilings are daily/monthly only; weekly and lifetime ceilings exist for API keys but not as named policies.
  • Guardrails apply to text you send; they do not inspect provider replies.
  • Disabling or deleting a policy is itself audited with the reason you give.

For the money side these restrictions protect, see Wallets and financial audit.