generated: '2026-08-11'
method: searched
source: >-
  https://alphai.io/developers and https://alphai.io/llms-full.txt, corroborated
  by the OpenAPI info.description and the RateLimited response component in
  openapi/alphaai-rest-api-openapi.yml.
description: >-
  A two-layer quota model — a per-minute burst cap and a per-day volume cap —
  where a request passes only if both are under budget. The published runtime
  signal reports the DAY layer only; the minute layer is invisible until it
  trips, which is the one thing an agent integrating this API has to know.

limit_count: 6
scope: per-account (all keys on an account share the budget)

limits:
  - tier: Free
    scope: per-account
    window: 1m
    limit: 20
  - tier: Free
    scope: per-account
    window: 1d
    limit: 100
    reset: 00:00 UTC
  - tier: Basic
    scope: per-account
    window: 1m
    limit: 60
  - tier: Basic
    scope: per-account
    window: 1d
    limit: 10000
    reset: 00:00 UTC
  - tier: Pro
    scope: per-account
    window: 1m
    limit: 150
  - tier: Pro
    scope: per-account
    window: 1d
    limit: 100000
    reset: 00:00 UTC

headers:
  - name: X-RateLimit-Limit
    on: every keyed response
    describes: the per-day volume cap for the caller's tier
  - name: X-RateLimit-Remaining
    on: every keyed response
    describes: requests left in the current UTC day
  - name: X-RateLimit-Reset
    on: every keyed response
    describes: epoch seconds of the next 00:00 UTC reset
  - name: Retry-After
    on: 429 only
    describes: seconds to wait; short when the per-minute burst cap tripped
style: X-RateLimit-* (legacy form, not the RFC 9331 RateLimit-* draft headers)

exhaustion:
  status: 429
  body_fields:
    - message
    - extra.tier
    - extra.limit_per_minute
    - extra.limit_per_day
    - extra.retry_after_seconds
    - extra.upgrade
  upgrade_block: >-
    Callers below Pro receive an extra.upgrade object carrying the higher tiers'
    caps and a pricing_url — the 429 is a self-describing upsell, which is a
    genuinely agent-friendly touch.

caveats:
  - >-
    The published headers describe ONLY the per-day layer. The per-minute burst
    cap has no header at all and surfaces solely as a 429 with a short
    Retry-After, so an agent watching X-RateLimit-Remaining can still be
    throttled without warning. This is documented by the provider, not inferred.
  - >-
    The docs note the headers reflect origin processing, so a response served
    from a shared cache may omit them entirely.
  - >-
    A separate 403 (not 429) enforces the archive horizon: paging past the
    tier's window returns extra.reason = archive_horizon. It is a quota, but it
    is signalled on a different status code.

related:
  plans: plans/alphaai-plans-pricing.yml
  conventions: conventions/alphaai-conventions.yml
  errors: errors/alphaai-problem-types.yml