generated: '2026-08-13'
method: searched
source: https://docs.lucidya.com/docs/Social-Listening-api/wi0q2azbypqk5-rate-limiting
name: Lucidya API Rate Limits
type: RateLimits
summary: >-
  Lucidya publishes a per-product Rate Limiting article on every API. Limits are
  plan-dependent floors ("start from"), enforced per minute with a hard reset at
  the top of the next minute. No rate-limit response headers are published or
  declared in any of the five OpenAPI documents, so the runtime signal an agent
  needs does not exist — only the number in the docs.
limit_count: 4
limits:
  - id: rest-default
    scope: per-key
    applies_to:
      - Social Listening API
      - CDP API
      - OmniChannel API
      - OmniServe Analytics API
    window: 1 minute
    limit: 100
    unit: requests
    floor: true
    note: >-
      "Our rate limits start from 100 requests per minute." The value is the
      entry-level floor; the effective limit depends on the subscribed plan.
    docs:
      - https://docs.lucidya.com/docs/Social-Listening-api/wi0q2azbypqk5-rate-limiting
      - https://docs.lucidya.com/docs/cdp-api/u5a4p6o6egebt-rate-limiting
      - https://docs.lucidya.com/docs/omnichannel-api/zrzb7o6iwt1ho-rate-limiting
  - id: ai-requests
    scope: per-key
    applies_to:
      - AI API
    window: 1 minute
    limit: 6
    unit: requests
    floor: true
    note: >-
      "Our rate limits for AI API start from 6 requests per minute." Materially
      tighter than the 100/min applied to the other four products — an agent
      built against the general figure will be throttled on the AI surface.
    docs: https://docs.lucidya.com/docs/ai-api/12m2bpnec7ycv-rate-limiting
  - id: ai-batch-size
    scope: per-request
    applies_to:
      - AI API
    window: null
    limit: 100
    unit: texts per request
    floor: true
    note: >-
      "100 texts (Batch Size) per request (It can be changed when creating
      package)." Applies to the four batch prediction operations
      (predict_sentiment_batch, predict_dialects_batch, predict_domains_batch,
      predict_themes_batch).
    docs: https://docs.lucidya.com/docs/ai-api/12m2bpnec7ycv-rate-limiting
  - id: ai-audio-duration
    scope: per-key
    applies_to:
      - AI API (Audio Transcription)
    window: 1 minute
    limit: 100
    unit: minutes of audio per minute
    note: >-
      "For Audio Analysis, the total duration of one audio file or multiple audio
      files should not exceed 100 minute per minute." A duration quota rather
      than a request-count quota.
    docs: https://docs.lucidya.com/docs/ai-api/12m2bpnec7ycv-rate-limiting
exhaustion:
  status: 429
  slug: too_many_requests
  body: '{"error":{"status":429,"detail":"..."}}'
  reset: >-
    "Subsequent requests will be denied until the start of the next minute" —
    a fixed-window reset on the minute boundary, not a rolling window.
response_headers:
  published: false
  observed: false
  checked:
    - RateLimit-Limit / RateLimit-Remaining / RateLimit-Reset
    - X-RateLimit-Limit / X-RateLimit-Remaining / X-RateLimit-Reset
    - Retry-After
  detail: >-
    None of the above appear in the Rate Limiting articles, in the Responses
    error registry, or as declared response headers on any of the 73 operations
    in the five OpenAPI documents. Live header observation is not possible: every
    api.lucidya.com route requires a CSM-issued token, and unauthenticated calls
    are answered by Cloudflare Access or a Kong "no Route matched" 404 before any
    rate-limit middleware runs.
  impact: >-
    A client cannot read remaining budget, cannot learn its actual plan ceiling
    at runtime, and gets no Retry-After on 429. The only correct strategy is to
    self-pace below the documented floor and back off to the next minute boundary
    on 429. This is the single highest-value runtime fix available to Lucidya:
    emitting RateLimit-* and Retry-After would turn four prose numbers into a
    machine signal.
related_quotas:
  note: >-
    Commercial quotas that are not per-minute API rate limits but do bound API
    usage are recorded in ../plans/lucidya-ltd-plans-pricing.yml — notably the
    OmniServe reply caps (150 / 800 on Growth, 400 / 2,000 on Enterprise), the
    12-month data retention, survey-response caps, and the per-seat AI credit
    pools.
  blocking_behaviour: >-
    Separately from 429, the published error registry states that a caller
    "blocked due to many errors in particular time" receives 403 forbidden. A
    client that hammers on 4xx can therefore be blocked outright rather than
    throttled.
cross_links:
  errors: ../errors/lucidya-ltd-problem-types.yml
  conventions: ../conventions/lucidya-ltd-conventions.yml
  plans: ../plans/lucidya-ltd-plans-pricing.yml