generated: '2026-09-06'
method: searched
source: >-
  https://apis.daglo.ai/openapi.prod.yaml (info.description, "API Rate Limits" section) and
  https://developers.daglo.ai/guide/en/STT-Async.html
limit_count: 1
limits:
  - scope: per-endpoint
    window: 1 second
    limit: 20
    unit: requests
    burst: null
    applies_to: every API endpoint
    exhaustion_status: 429
    published_text: >-
      '각 API Endpoint 별 20 requests/sec 의 Rate Limit이 적용됩니다.' — a rate limit of 20 requests/sec
      applies per API endpoint.
    enforcement_note: >-
      The contract adds that repeated, sustained over-limit traffic is treated as abnormal use and the
      requests may be blocked outright — so 429 is not necessarily the worst outcome of ignoring it.
    source: https://apis.daglo.ai/openapi.prod.yaml
response_headers:
  documented: false
  ratelimit_headers: []
  retry_after: false
  note: >-
    No RateLimit-*, X-RateLimit-* or Retry-After header appears in any response definition in either
    OpenAPI document, and none is described in the guide. An agent gets no runtime budget signal at all:
    it cannot see how much of the 20/sec allowance it has used, and on a 429 it has no server-supplied
    interval to wait, so back-off has to be guessed. This is the single clearest agent-readiness gap in
    an otherwise tidy contract.
  probe_note: >-
    Not probed against a live 429 — provoking one would require an authenticated token and deliberately
    exceeding a production limit, which this pipeline does not do.
related_capacity_limits:
  - subject: async STT upload
    limit: 2GB file size, 4 hours duration
    source: https://developers.daglo.ai/guide/en/STT-Async.html
  - subject: sync STT upload
    limit: 30 seconds of audio
    source: https://developers.daglo.ai/guide/en/FAQ.html
  - subject: realtime gRPC stream
    limit: 6 hours per request stream
    source: https://developers.daglo.ai/guide/Quick-Realtime-Voice-To-Text.html
  - subject: realtime gRPC audio format
    limit: LINEAR16, 16000 Hz, mono (1 channel) only
    source: grpc/actionpower-speech.proto
  - subject: multi-channel STT
    limit: at most 2 channels
    source: openapi/actionpower-daglo-cloud-api-openapi.yml
quota_exhaustion:
  behaviour: >-
    Separate from the rate limit: when an account's credits run out the request is interrupted rather
    than throttled, and returns 403 rather than 429. See sandbox/actionpower-sandbox.yml.