generated: '2026-08-09'
method: searched
source: openapi/linkagi-model-api-openapi.json
docs: https://docs.linktoagi.com/
note: >-
  LinkAGI is a protocol-imitating relay: its request/response conventions are whatever the upstream vendor
  protocol defines for the route you choose. There is no LinkAGI-native envelope, no LinkAGI-native
  pagination, and no idempotency contract. What IS LinkAGI-specific is the auth header per route, the token
  "group" (号池) that decides which models a key can see, and the gateway's error/request-id shape.
authentication:
  style: per-route static API key issued in the console; no OAuth, no OIDC, no refresh
  routes:
  - route: /v1/models, /v1/chat/completions, /v1/responses
    scheme: bearerAuth
    header: 'Authorization: Bearer <key>'
  - route: /v1/messages
    scheme: anthropicApiKey
    header: 'x-api-key: <key>'
    extra_required_header: 'anthropic-version: 2023-06-01'
  - route: /v1beta/models/{model}:generateContent
    scheme: geminiApiKey
    header: 'x-goog-api-key: <key>'
  grouping:
    concept: token group / 号池
    description: >-
      A key belongs to a group; the group determines which models are visible and at what ratio. Group
      names are published unauthenticated at GET /api/pricing under usable_group and group_ratio (e.g.
      "Codex | Pro号池", "ClaudeCode-补贴渠道", "Gemini", "default" = 按次计费).
    consequence: A 401, a 400 "model not found", or an empty model list is frequently a group problem, not
      a key problem.
  see_also: authentication/linkagi-model-api-authentication.yml
idempotency:
  supported: false
  header: null
  note: No idempotency key header or parameter appears in the OpenAPI or the documentation. Retries after a
    504 may be double-billed; the docs advise bounding retries rather than offering an idempotency contract.
pagination:
  supported: false
  note: No collection endpoint paginates. GET /v1/models returns a single unpaged list, and GET /api/pricing
    returns the whole model array (58 models on 2026-08-09).
field_expansion:
  supported: false
metadata:
  supported: false
request_tracing:
  header: x-oneapi-request-id
  also_in_body: 'error.message, appended as "(request id: ...)"'
  note: Emitted by the New API gateway on both success and error responses; quote it when contacting support.
versioning:
  style: uri-path inherited from the imitated vendor protocol (/v1, /v1beta)
  see_also: lifecycle/linkagi-model-api-lifecycle.yml
error_envelope:
  shape: '{"error": {"code": "", "message": "...", "type": "new_api_error"}}'
  content_type: application/json; charset=utf-8
  rfc9457: false
  see_also: errors/linkagi-model-api-problem-types.yml
rate_limit_signaling:
  status: 429
  headers_published: false
  numeric_limits_published: false
  guidance: exponential backoff plus a concurrency cap; identify which layer (client, group, model pool,
    upstream vendor) imposed the limit from the console call log
  docs: https://docs.linktoagi.com/openai-api-429-rate-limit.html
streaming:
  parameter: stream (boolean) on ChatCompletionRequest and ResponseRequest
  verified: false
  note: The provider's own evidence boundary lists streaming as not verified.
billing:
  model: prepaid pay-as-you-go, CNY
  currency_symbol: '¥'
  unit: quota_per_unit 500000, quota_display_type CNY (from GET /api/status)
  pricing_api: https://api.linktoagi.com/api/pricing
  pricing_version_field: pricing_version (hash; changes when the price table changes)
  guidance: retain the pricing_version or the collection timestamp alongside any quoted price, and reconcile
    paid calls against the console usage log
client_onboarding:
  codex:
    base_url: https://api.linktoagi.com/v1
    protocol: Responses
    docs: https://docs.linktoagi.com/codex-api.html
  claude_code:
    env: ANTHROPIC_BASE_URL=https://api.linktoagi.com
    route: /v1/messages
    docs: https://docs.linktoagi.com/claude-code-api.html
  gemini_cli:
    env: GOOGLE_GEMINI_BASE_URL=https://api.linktoagi.com
    route: /v1beta/models/...
    docs: https://docs.linktoagi.com/gemini-cli-api.html