generated: '2026-07-27'
method: searched
source:
- https://useapi.net/llms.txt
- https://useapi.net/docs/start-here/setup-useapi
- https://useapi.net/docs/account-management
- https://useapi.net/docs/questions-and-answers
- https://useapi.net/docs/api-google-flow-v1
- openapi/ (derived from the first-party Postman collections)
summary: >-
  useapi.net publishes exactly one platform-wide convention — bearer-token authentication — and is
  explicit in its own documentation that everything else is per-service. The vendor states:
  "Identifier names (jobid / taskId / musicId / etc.), job lifecycle, response shapes, webhook
  payload shape, status code semantics, and synchronous-vs-asynchronous behavior vary per API."
  Integrators must treat each of the ten APIs as a separate contract; this artifact records what is
  genuinely cross-cutting and flags where the platform is deliberately heterogeneous.

authentication:
  style: bearer
  header: 'Authorization: Bearer user:<number>-<unique-string>'
  token_format: 'user:<number>-<unique-string>'
  scope: >-
    A single token authorizes every API under the subscription. There is no OAuth, no per-API key,
    and no scope system.
  rules:
  - Use the complete token string including the `user:` prefix and the alphanumeric suffix.
  - Do not truncate to just the numeric portion.
  - Do not URL-encode the token.
  delivery: Token is emailed on subscription; see https://useapi.net/docs/start-here/setup-useapi
  artifact: authentication/useapi-authentication.yml

bring_your_own_account:
  model: >-
    useapi.net is a fronting layer, not a first-party model host. Before calling a generation
    endpoint the caller registers one or more of their own accounts on the underlying AI service
    (via POST /<service>/accounts, supplying browser cookies or a session token). An active
    subscription on the underlying service is required.
  load_balancing: >-
    When the per-request account selector (`email` / `account` / `channel_id`, depending on the
    service) is omitted, useapi.net picks an account automatically using a weighted health score
    built from executing / completed / failed / rateLimited counts over the last 15 minutes.
  quarantine: >-
    Accounts that return upstream 429s are pulled from rotation automatically. On Google Flow the
    documented cooldowns are ~30 minutes for USER_QUOTA_REACHED and USER_REQUESTS_THROTTLED, and
    until UTC midnight for PER_MODEL_DAILY_QUOTA_REACHED. Quarantine is skipped when exactly one
    account is configured.

idempotency:
  supported: false
  evidence: >-
    No idempotency key header or parameter is documented on any endpoint, and none appears in any
    of the ten first-party Postman collections. The only use of the word "idempotent" in the
    documentation describes an internal implementation detail (repeated 429s for the same
    account/reason/model tuple preserve the original quarantine firstSeen timestamp) and is not a
    caller-facing contract.
  practical_guidance: >-
    Generation calls are not safe to blind-retry — a retried POST creates a second job and consumes
    credits on the underlying account. Use the returned job identifier and the per-service job
    endpoint (or a replyUrl webhook) to determine outcome before retrying.

asynchrony:
  model: job-based
  pattern: >-
    Most generation endpoints create a job and return an identifier; the caller then polls the
    service's job endpoint or receives a replyUrl webhook callback.
  identifier_names_vary: true
  examples:
  - service: google-flow
    identifier: jobid
    poll: GET /v1/google-flow/jobs/{jobId}
  - service: dreamina
    identifier: jobid
    poll: GET /v1/dreamina/videos/{jobid}
  - service: flowmusic
    identifier: jobid
    poll: GET /v1/flowmusic/jobs/{jobid}
    note: Synchronous by default; async is opt-in via an `async` flag plus replyUrl.
  - service: mureka
    identifier: musicId
  - service: pixverse
    identifier: taskId

webhooks:
  supported: true
  parameter: replyUrl
  set_globally: 'POST https://api.useapi.net/v2/account with {"replyUrl": "..."} sets the default for every API'
  override: Any individual call may override the account-level default with its own replyUrl.
  fires_on: job completed or failed
  payload_shape: per-service — the vendor explicitly states webhook payload shape varies by API
  artifact: asyncapi/useapi-jobs-webhooks.yml

pagination:
  standardized: false
  note: >-
    There is no platform-wide pagination contract; each service uses the paging vocabulary of the
    site it wraps. Observed parameters across the ten first-party Postman collections:
  styles:
  - style: limit-offset
    params: [limit, offset]
  - style: cursor
    params: [cursor]
  - style: last-id / keyset
    params: [last_id, lastId, lastVideoId, lastImageId, lastMusicId, lastChatID]
  - style: page-number
    params: [page, pageSize, page_size, pageNum]
  account_stats:
    endpoint: GET https://api.useapi.net/v2/account/stats
    params: [bot, date, limit, config]
    max_limit: 50000
    note: When `limit` is supplied the `date` filter is ignored and the last N records are returned.

versioning:
  scheme: uri-path
  pattern: https://api.useapi.net/{version}/{service}
  observed_versions: [v1, v2]
  note: >-
    The version is per-service, not per-platform — PixVerse is on v2 while the other current
    services are on v1, and the cross-cutting account API is on v2. Version bumps are announced on
    the blog; the retired Midjourney surface is the one documented precedent for a v1 to v2 cutover.
  artifact: lifecycle/useapi-lifecycle.yml

error_envelope:
  standardized: false
  problem_json: false
  shapes:
  - shape: flat
    example: '{"error": "Unauthorized"}'
    note: useapi.net platform-level errors (auth, account not found, validation).
  - shape: passthrough
    example: '{"error": {"code": 400, "message": "...", "status": "INVALID_ARGUMENT", "details": [...]}}'
    note: >-
      Errors originating at the wrapped provider are forwarded largely verbatim, so the envelope is
      the upstream vendor's (the example above is Google's google.rpc.ErrorInfo shape).
  artifact: errors/useapi-problem-types.yml

rate_limits:
  publishes_quotas: false
  signaling:
    status: 429
    headers: [Retry-After]
    body_field: retryAfter
    body_field_format: ISO 8601 timestamp
  note: >-
    useapi.net does not impose or publish its own request quota; 429s originate from the wrapped
    provider (or from useapi.net's own `no_eligible_account` short-circuit once every linked
    account is quarantined). Every 429, including the short-circuit, carries a Retry-After header
    and a retryAfter body field, and callers are told to honor it.
  reason_codes: >-
    On Google Flow the 429 is further qualified by an upstream reason —
    PUBLIC_ERROR_UNUSUAL_ACTIVITY_TOO_MUCH_TRAFFIC, PUBLIC_ERROR_USER_REQUESTS_THROTTLED,
    PUBLIC_ERROR_PER_MODEL_DAILY_QUOTA_REACHED, PUBLIC_ERROR_USER_QUOTA_REACHED — each with a
    different remediation and cooldown. See errors/useapi-problem-types.yml.

request_tracing:
  request_id_header: none documented
  observability: >-
    GET https://api.useapi.net/v2/account/stats is the audit surface — per-bot request counts broken
    down by config, endpoint, model, HTTP status, status text, tier and SKU, with average latency
    and success rate. Data latency is 5-15 minutes and retention is 3 months.

content_moderation:
  prevalidation: false
  statement: >-
    "Our API does not perform any prompt pre-validations — it will return an error status code when
    the underlying AI service moderates the job." The vendor recommends caller-side screening and
    points to its own free MiniMax-Text-01 endpoint (POST /v1/minimax/llm) for text and image safety
    checks.

media_upload:
  pattern: >-
    Binary assets are POSTed as a raw body with the file's own Content-Type (not multipart), e.g.
    POST /v1/runwayml/assets/?name=<name> and POST /v1/minimax/files/.

cross_service_caveat: >-
  The same underlying model (for example Seedance 2.0, Veo 3.1 or Nano Banana) is frequently exposed
  by more than one useapi.net API, but the request and response shapes differ between them — each
  API exposes the native interface of the site it wraps. https://useapi.net/model-matrix is the
  canonical map of which model is available through which API.