generated: '2026-08-13'
method: searched
source: https://api.signal-ai.com/docs
source_detail: >-
  Published verbatim in the "Rate Limiting" section of the Signal AI API
  reference, which is rendered from info.description of
  https://api.signal-ai.com/openapi.json (fetched 2026-08-13, HTTP 200).
scope_model: per-endpoint, per-Client ID
limit_count: 4
limits:
- name: Content search
  scope: per-endpoint-per-client-id
  endpoints:
  - POST /search
  window: second
  limit: 2
- name: Content search
  scope: per-endpoint-per-client-id
  endpoints:
  - POST /search
  window: minute
  limit: 30
- name: Content metrics
  scope: per-endpoint-per-client-id
  endpoints:
  - POST /metrics
  window: second
  limit: 2
- name: Content metrics
  scope: per-endpoint-per-client-id
  endpoints:
  - POST /metrics
  window: minute
  limit: 15
- name: Concept affinity
  scope: per-endpoint-per-client-id
  endpoints:
  - POST /affinity
  window: second
  limit: 2
- name: Concept affinity
  scope: per-endpoint-per-client-id
  endpoints:
  - POST /affinity
  window: minute
  limit: 30
- name: All others
  scope: per-endpoint-per-client-id
  endpoints:
  - GET /topics
  - GET /topics/{id}
  - GET /entities
  - GET /entities/{id}
  - GET /sources
  - GET /sources/{id}
  - GET /source-locations
  - GET /categories/iptc-media-topics
  - POST /events
  - GET /events/{hash}
  - POST /risk-events-search
  - POST /risk-events-scores
  - GET /risk-events-definitions
  - GET /users
  window: second
  limit: 5
- name: All others
  scope: per-endpoint-per-client-id
  endpoints:
  - all endpoints not listed above
  window: minute
  limit: 60
exhaustion:
  status_code: 429
  documented_as: >-
    "429 – Too many requests: The server received too many requests in the last
    second or minute."
  retry_guidance: >-
    Not stated in the reference. The first-party `signal-api-tools` client treats
    429 as a retryable (non-fatal) status and backs off.
response_headers:
  documented: false
  observed: []
  notes: >-
    Signal AI documents NO rate-limit response headers — no `X-RateLimit-*`, no
    RFC 9331 `RateLimit-*`, no `Retry-After`. An unauthenticated
    GET https://api.signal-ai.com/topics (probed 2026-08-13, HTTP 401) returned
    only `signal-version: 1` plus standard Cloudflare/security headers, so no
    runtime budget signal could be observed anonymously either. A client cannot
    see how much of its per-second/per-minute allowance remains; it can only
    count its own calls against the published table or wait for a 429. This is
    the gap between a documented limit and an agent-usable one.