Documentation
Rate limiting

Gate Rate Limiting

Gate/AI enforces limits at the edge, before a request reaches your provider key. A gate-wide ceiling protects your bill from spikes; per-device budgets stop one user from spending everyone's allowance; tier overrides turn "you've hit your limit" into an upgrade moment.

Analytics guide iOS SDK Android SDK

Two kinds of limits

Gate rate limit

Core and Pro

A hard ceiling on requests per minute across every device using the gate. It's the circuit breaker: a scripted client, a runaway retry loop, or a viral moment can't run up your provider bill past the number you chose.

Configure it from the gate page → Gate node → Gate rate limit. Blank means no cap (system ceilings still apply).

Device usage limits

Pro

A budget each device gets over a reset period — requests and/or tokens — plus an optional daily burst cap, with overrides per user tier. This is the limit your users feel, so it comes with usage headers and informative errors your app can turn into meters and upgrade prompts.

Configure it from the gate page → Quotas node.

Setting up device limits

The Quotas panel walks through the policy top to bottom and restates it in English before you save.

  1. 1

    Pick a reset period

    Calendar month resets for everyone on the 1st (UTC) and is the easiest to explain. Rolling 30 days has no cliff at month end but is harder to show in a meter. User's billing cycle resets on each user's renewal day and lines up with your paywall — it needs the app to report the renewal day (see below). A device is measured against one period; pick the one that matches how you sell.

  2. 2

    Set the budget per device

    Requests and tokens for that period. Turn a metric off to leave it uncapped. Presets are there so you don't count zeros.

  3. 3

    Optionally cap each day

    A burst limit on top of the budget so one device can't spend a month's allowance in an afternoon. Resets at midnight UTC.

  4. 4

    Add tier overrides

    For each userTier your app sends, choose per metric: same as default, a custom amount, or unlimited. Names must match the client value exactly (case-sensitive).

  5. 5

    Save

    Limits take effect on the next request. The strip on the gate page shows the policy and how many devices were throttled today.

How enforcement counts. The request that crosses a quota is allowed; requests after it are rejected until the window resets. Rejected requests consume no budget. All windows are UTC. A device is identified by the SDK's device identifier, falling back to IP address when it isn't sent.

Billing-cycle periods

To measure a device against the user's own subscription month, the app reports the day of month their subscription renews (from StoreKit, Play Billing, or RevenueCat). Days 29–31 clamp to short months automatically. Requires SDK 1.1.0+. Devices that never report a day still accrue usage from their first request, so nobody is exempt.

Swift

gateAIClient.quotaAnchorDay = renewalDayOfMonth  // 1–31

Kotlin

gateAIClient.quotaAnchorDay = renewalDayOfMonth  // 1–31

Tiers: the cap as your paywall

Set a modest default budget, then give paying tiers more — or no limit at all. When a free user hits the cap, the 429 tells your app exactly which window closed and when it reopens, which is the moment to show the upgrade screen.

Example policy Requests / month Tokens / month
Default (no tier, or an unknown tier) 200 100,000
pro 5,000 Unlimited
team Same as default 2,000,000

The app sets gateAIClient.userTier = "pro" after a purchase (see the analytics guide). Tier is an enforcement key, so keep names stable; use userStatus for anything free-form.

What your app sees

Quota headers on every response

Remaining budget for the tightest configured window, per metric, so you can render a meter without a separate call:

  • X-Quota-Requests-Remaining · X-Quota-Requests-Limit · X-Quota-Requests-Reset
  • X-Quota-Tokens-Remaining · X-Quota-Tokens-Limit · X-Quota-Tokens-Reset

The SDKs parse these into GateAIQuotaStatus (Swift) / QuotaStatus (Kotlin) with used-fraction helpers.

A structured 429 when a limit is hit

{ "error": "rate_limited",
  "code": "device_monthly_requests_exceeded",
  "window": "monthly", "limit": 200, "used": 200,
  "resets_at": "2026-10-01T00:00:00.000Z" }

Exposed as GateAIError.rateLimitInfo / GateApiException.rateLimitInfo.

Code Meaning
device_daily_requests_exceeded / device_daily_tokens_exceeded The daily burst cap for this device.
device_monthly_requests_exceeded / device_monthly_tokens_exceeded Calendar-month budget.
device_rolling_30d_requests_exceeded / device_rolling_30d_tokens_exceeded Rolling 30-day budget.
device_cycle_requests_exceeded / device_cycle_tokens_exceeded User's billing-cycle budget.
gate_requests_per_minute_exceeded The gate-wide ceiling; nothing to do with this particular device.
account_unique_clients_exceeded Your plan's monthly unique-device allowance is used up (Free: 25).
global_rpm_exceeded A platform-wide safety ceiling; contact support if you see this.

Rate-limited requests appear in Logs with an RL badge and in the Analytics tab's rate-limit offenders, and the strip on the gate page counts devices throttled today.