generated: '2026-08-14'
method: searched
source: >-
  https://api.voygr.tech/openapi.json (info.description "## Limits" and the 429
  response description on POST /calls),
  https://github.com/voygr-tech/callwright-skill (README.md, SKILL.md)
docs: https://api.voygr.tech/docs
limit_count: 5
rate_limits:
- id: free-per-second
  scope: per-key
  tier: free
  window: 1s
  limit: 5
  unit: requests
  status_on_exhaustion: 429
  error_code: RATE_LIMIT_ERROR
- id: free-per-minute
  scope: per-key
  tier: free
  window: 60s
  limit: 10
  unit: requests
  status_on_exhaustion: 429
  error_code: RATE_LIMIT_ERROR
- id: paid-per-second
  scope: per-key
  tier: paid
  window: 1s
  limit: 10
  unit: requests
  status_on_exhaustion: 429
  error_code: RATE_LIMIT_ERROR
- id: paid-per-minute
  scope: per-key
  tier: paid
  window: 60s
  limit: 100
  unit: requests
  status_on_exhaustion: 429
  error_code: RATE_LIMIT_ERROR
- id: concurrent-calls
  scope: per-key
  window: concurrent
  limit: 2
  unit: calls in flight
  note: >-
    "typically 2" per the spec; the exact cap is returned per key as
    max_concurrent_calls on GET /users/me.
  status_on_exhaustion: 409
  response_field: active_call_ids
daily_call_caps:
- tier: free
  limit: 25
  unit: calls/day
  source: https://github.com/voygr-tech/callwright-skill README + SKILL.md
- tier: paid
  limit: 5000
  unit: calls/day
  note: Any credit purchase lifts the cap; credits then become the practical limit.
response_headers:
  observed: []
  probed:
  - url: https://api.voygr.tech/health
    status: 200
    headers_present:
    - date
    - content-type
    - content-length
    - set-cookie
    - server
  - url: https://api.voygr.tech/users/me
    status: 401
    headers_present:
    - date
    - content-type
    - content-length
    - set-cookie
    - server
  finding: >-
    NO runtime rate-limit signal. Live responses carry no RateLimit-*, no
    X-RateLimit-*, and no Retry-After. The strings "RateLimit", "Retry-After"
    and "retry_after" appear nowhere in the published OpenAPI. An agent cannot
    learn its remaining budget from a response — it must either read the numbers
    out of prose documentation or poll GET /v1/usage / GET /users/me. This is
    the largest agent-readiness gap on the runtime surface.
    (Probed 2026-08-14; `server: voygr-gateway` behind an AWS ALB.)
quota_endpoints:
- operation: get_usage_v1_usage_get
  path: GET /v1/usage
  returns:
  - available
  - call_credit_hold
  - quota_limit
  - current_usage
  - remaining
  - percentage_used
  - tier
  note: '`available // call_credit_hold` is how many calls can be started.'
- operation: get_me
  path: GET /users/me
  returns:
  - customer_id
  - quota_limit
  - current_usage
  - credits_available
  - credits_held
  - max_concurrent_calls
maintenance:
  status: 503
  field: resume_at
  note: Service maintenance windows return 503 with a resume_at timestamp.
client_retry_behavior:
  provider_published: true
  source: https://github.com/voygr-tech/dev-tools
  detail: >-
    The first-party CLI always retries 429 and 5xx with exponential backoff, up
    to 3 attempts. The Python library ships with retries=0 by default.