generated: '2026-08-11'
method: searched
source: >-
  https://infer.flow7.org/openapi-public.json (429 response + Retry-After header),
  https://infer.flow7.org/docs, and https://infer.flow7.org/notes/one-cent-concurrent-spend-ceiling
  (Infer's own dated observation of ceiling enforcement).
description: >-
  Infer publishes NO numeric request-rate limit. What it does publish, in detail, is a second and
  more consequential limiter — a per-API-key spend ceiling, enforced against a pre-flight reservation
  before a request is admitted. Both surface as HTTP 429, which is the trap here: a client cannot tell
  "slow down" from "out of budget" without reading error.code.
provider: Infer by Flow7
providerId: infer-by-flow7
limit_count: 0
limit_count_note: >-
  Zero PUBLISHED numeric request-rate limits. This is an honest zero, not a missing probe: the OpenAPI
  declares a 429 on all four operations and a Retry-After header, and the docs describe throttling in
  prose, but no requests-per-second, per-minute or per-day figure is stated anywhere on the public
  surface. Unauthenticated GETs of /api/public/status and /api/public/catalog returned no RateLimit-*
  or X-RateLimit-* headers at all.
headers:
  observed_on_success:
    - name: X-Request-ID
      note: >-
        Probed live on GET /api/public/status (HTTP 200, 2026-08-11) — returned as x-request-id, e.g.
        req_6c40bad6e6d044e6b58ce18f. No rate-limit headers accompanied it.
  documented_on_429:
    - name: Retry-After
      type: integer seconds
      scope: coarse request-rate limits only
      source: components.responses.RateLimited.headers in the OpenAPI
      note: Explicitly NOT promised for spend-ceiling rejections.
    - name: X-Request-ID
      type: string
  absent:
    - RateLimit-Limit
    - RateLimit-Remaining
    - RateLimit-Reset
    - X-RateLimit-Limit
    - X-RateLimit-Remaining
    - X-RateLimit-Reset
  absence_note: >-
    No IETF draft RateLimit-* fields and no X-RateLimit-* fields on any observed response. An agent
    therefore has no runtime budget signal from headers and must either back off blindly on 429 or
    poll its own ledger.
response_codes:
  throttled: 429
  budget_exhausted: 429
  wallet_insufficient: 402
  capacity_exhausted: 503
limits: []
spend_limits:
  - name: Per-API-key daily spend ceiling
    scope: api-key
    metric: microdollars
    window: day
    limit: null
    limit_note: >-
      Customer-configurable; no default is published. Infer's Route Note 005 records a deliberately
      set $0.010000 (10,000 microdollar) daily ceiling on a temporary key.
    enforcement: >-
      Checked BEFORE a request is admitted, against the full reserved maximum — not against the
      settled charge. One active reservation equal to the ceiling blocks every concurrent request.
    error_code: daily_budget_exceeded
    status: 429
    observed: >-
      2026-08-11 18:33 UTC — one admitted request (HTTP 200) and seven concurrent contenders all
      rejected HTTP 429 daily_budget_exceeded within 553.380-556.326 ms, dispatch spread 0.638 ms.
      The admitted request settled at 43 microdollars and 9,957 microdollars were returned to the
      wallet.
    source: https://infer.flow7.org/notes/one-cent-concurrent-spend-ceiling
  - name: Per-API-key monthly spend ceiling
    scope: api-key
    metric: microdollars
    window: month
    limit: null
    limit_note: Customer-configurable; the same note records a matching $0.010000 monthly ceiling.
    error_code: daily_budget_exceeded
    status: 429
    source: https://infer.flow7.org/notes/one-cent-concurrent-spend-ceiling
  - name: Wallet balance
    scope: workspace
    metric: usd
    window: n/a
    limit: null
    enforcement: >-
      A reservation for the maximum estimated cost is taken before routing. If the wallet cannot cover
      it, the request is rejected 402 insufficient_credits before any supplier cost is incurred.
    error_code: insufficient_credits
    status: 402
    source: https://infer.flow7.org/docs
payload_limits:
  - name: Request body / input size
    status: 413
    limit: null
    note: >-
      "The input or request body exceeds a configured limit" — the limit is configured but not
      published as a number.
  - name: max_output_tokens
    limit: null
    note: >-
      Client-supplied ceiling, minimum 1. The OpenAPI states "The deployment's configured maximum also
      applies" without naming it.
client_guidance:
  - Read error.code on every 429 before backing off; Retry-After may be absent.
  - >-
    Concurrency against a low-ceiling key is effectively serialized, because the reservation — not the
    charge — is what consumes the ceiling.
  - >-
    GET /api/public/status before a paid request. Infer states repeatedly that a published catalog
    entry is not proof a selector is callable.
related:
  - conventions/infer-by-flow7-conventions.yml
  - errors/infer-by-flow7-problem-types.yml
  - plans/infer-by-flow7-plans-pricing.yml