generated: '2026-07-19'
method: searched
source: >-
  https://docs.flex.ai/inference-api/reference/openai-compatibility and
  https://docs.flex.ai/inference-api/reference/billing — cross-cutting
  request/response conventions of the FlexAI Token Factory OpenAI-compatible API.
description: >-
  How the FlexAI Token Factory API behaves across every operation:
  authentication, error envelope, rate-limit signaling, versioning, and
  streaming. FlexAI is a drop-in OpenAI-compatible API, so its conventions
  mirror the OpenAI API; FlexAI does not currently document an idempotency-key
  header.

base_url: https://tokens.flex.ai/v1
api_style: REST over HTTPS, JSON request/response, OpenAI-compatible

authentication:
  scheme: Bearer token (API key)
  header: 'Authorization: Bearer $FLEXAI_API_KEY'
  key_management: https://tokens.flex.ai/signup
  oauth: false
  detail: authentication/flexai-authentication.yml
  docs: https://docs.flex.ai/inference-api/reference/openai-compatibility

idempotency:
  supported: false
  note: >-
    FlexAI does not document an Idempotency-Key header. The API is
    OpenAI-compatible; a `seed` parameter is accepted on chat/completions for
    reproducible sampling but is not an idempotency mechanism. Retries are the
    client's responsibility.

pagination:
  supported: false
  note: >-
    The catalog endpoint (GET /v1/models) returns the full list in one
    response; no cursor/offset pagination is documented.

streaming:
  supported: true
  mechanism: 'Server-sent events when `stream: true` on chat/completions'
  usage: 'Set `stream_options.include_usage: true` to receive a final usage chunk.'
  note: >-
    Streaming responses are billed for all generated tokens even if the
    connection closes early.

versioning:
  scheme: uri-path
  current: v1
  note: The version is pinned in the base URL path (/v1); no version header.
  detail: lifecycle/flexai-lifecycle.yml

error_envelope:
  media_type: application/json
  rfc9457: false
  shape: '{ "error": { "message", "type", "param", "code", "doc_url" } }'
  note: >-
    OpenAI-compatible error object. `param` names the offending request field on
    validation errors; `authentication_error` responses include a `doc_url`
    extension.
  detail: errors/flexai-problem-types.yml
  docs: https://docs.flex.ai/inference-api/reference/openai-compatibility

rate_limits:
  signal_status: 429
  budget_status: 402
  headers:
    - x-ratelimit-limit-requests
    - x-ratelimit-remaining-requests
    - x-ratelimit-reset-requests
    - Retry-After
  detail: rate-limits/flexai-rate-limits.yml
  docs: https://docs.flex.ai/inference-api/reference/billing

output_limits:
  max_output_tokens_non_streaming: 2048
  note: >-
    Non-streaming chat/completions responses are capped at 2048 output tokens
    regardless of max_tokens (finish_reason "length"). `n` > 1 is rejected with
    a 400.

webhooks:
  supported: false
  note: No webhook or event-delivery surface is documented.