Usage Budgets
Steinkauz AI Budgets govern how much inference spend (money and/or tokens) each member or dedicated API key may use on each provider instance. They are optional policy ceilings, not included inference, and not a commercial seat meter.
Where to configure
- Settings → Budgets: organization owners and admins set optional caps per provider.
- Settings → API Keys: create keys in inherit (default) or dedicated Budget mode.
On Private Deployment installs there is no Stripe plan requirement. You can still open Settings → Budgets and set optional money or token caps; they reset on the UTC calendar month.
Optional BYOK caps
On each provider tab, admins set optional:
- Users: each member’s money and/or token caps
- Dedicated keys: caps on keys in dedicated Budget mode
Empty means uncapped; 0 means immediately exhausted. Edits that would set a cap below that subject’s spend this period are rejected.
API key modes
| Mode | Who is gated |
|---|---|
| Inherit (default) | Creator’s user Budgets (same as chat) |
| Dedicated | The key itself; optional BYOK / token caps (skip = uncapped) |
You can switch modes after create. Spend history stays on the subject that recorded it. Switching dedicated → inherit (or revoking a dedicated key) retires optional BYOK/token rows on the key.
Exhaustion
When any configured unit (money, input tokens, or output tokens) has Remaining ≤ 0, further inference for that subject on that provider is blocked. Chat hard-stops; Platform API returns 402 insufficient_quota on the next attempt after an overshooting call (that call still completes with HTTP 200).
Providers with no Budget row stay uncapped by Steinkauz AI until you set optional caps on the Budgets page.
A customer-owned Vercel AI Gateway connection still needs your Gateway API key.
Other providers (BYOK)
Each member can have optional Cost ($), Tokens In, and Tokens Out caps. Token caps show compact counts (K/M/B) until you focus the field. Period spend shows used/cap progress when a track is set.
If the provider reports tokens but not cost (or the reverse), a warning appears once for that provider: those Budgets cannot gate spend until metering matches. Saving never blocks on that warning.
Unsaved edits for the active provider use the shared save bar (Save applies all dirty member rows on that tab).
Configuration Audit
Cap edits appear in Settings → Configuration Audit (and as five recent events on the Budgets page). See Configuration Audit.