Skip to main content

​
What You Need in Advance

  • Account created and logged in.
  • At least one API key created in the console and associated with the correct API key group or team context.
  • The corresponding group has configured available models, channels, and basic quotas (see “Team Collaboration / Quotas and Usage” for details).

​
Model and Channel Limitations

  • Isolated by API Key:
    Each API key can only see and call models and channels configured under its own group; model lists for different API keys on the same platform may be completely different.
  • Model Directory is Authoritative:
    Lists returned by /v1/models, /claude/v1/models, /gemini/*/models represent models and sources truly available to the current API key; models not in the list are considered unavailable.
  • Multimodal and Task Channels:
    Task-based capabilities like Midjourney, Suno, RecraftAI, and Kling depend on corresponding channels being enabled. When not enabled, related task interfaces will be unavailable.

​
Quota and Billing Limitations

  • Insufficient Quota Rejects Immediately:
    During quota pre-deduction, if quota is insufficient on any side (personal, team, or owner), the request will be directly rejected and will not enter the model execution phase.
  • Pre-deduction and Settlement Two-Phase:
    Quota is pre-deducted when a request is made, and actual billing occurs after request completion based on real usage; failed requests do not consume additional quota.
  • Independent Quota Spaces:
    Quotas for different API keys/teams/contexts are independent and do not interfere with each other; when switching contexts, confirm which quota space is currently being used.

​
Protocol and Entry Point Limitations

  • OpenAI-Compatible Entry Point:
    /v1/* series interfaces only accept OpenAI-style request bodies and fields; if using Claude or Gemini official SDKs, prefer their respective native paths.
  • Native Paths:
    Paths like /claude/v1/messages, /gemini/:version/models/:model maintain their respective protocol rules; fields and model IDs are compatible with official versions, but successful calls still depend on API key-visible model limitations.
  • Task Interfaces:
    Interfaces for Midjourney, Suno, RecraftAI, Kling, etc. use an asynchronous “task ID + query” pattern and do not guarantee returning final images, audio, or video in a single request.

​
Rate and Stability Limitations

  • Rate Limiting and 429:
    To protect stability, the platform applies rate limiting to main routes and task interfaces based on configuration; when exceeded, 429 or related errors may be returned, requiring client-side retry or backoff.
  • Channel Health:
    Automatic retry and channel switching depend on at least one healthy backup channel; when all related channels are unavailable, requests may still fail.
  • Long Streaming Calls:
    Streaming responses will be closed on network exceptions or timeouts; reserve timeout and reconnection strategies on the client side.

​
Permission and Context Limitations

  • Context Types:
    Calls execute under personal (user) or team (team) contexts, with different contexts corresponding to different quotas and model visibility ranges.
  • Roles and Permissions:
    Whether you can manage team quotas, view team usage, or adjust member quotas depends on team roles and permission configuration (see “Team Collaboration / Roles and Permissions” for details).
  • Log Visibility Scope:
    Logs and usage statistics control visibility by context and permissions; regular members cannot view call records outside their permission scope.

​
Relationship with Other Capabilities

  • Quotas and Usage:
    All calls (including multimodal and tasks) are reflected in usage and quota views; quota interception logic is consistent with quota rules in team collaboration.
  • Security and Notifications:
    Content moderation and alert notifications depend on corresponding configuration in the console; if not configured, they are only reflected through returned errors and console logs.

​
Pre-Use Checklist

  1. In the console, confirm the current API key’s group, context (personal/team), and basic quota.
  2. Through the model directory interface, confirm whether the target model is in the current API key’s visible list.
  3. If using task interfaces like Midjourney, Suno, RecraftAI, Kling, first confirm that corresponding channels have been enabled by administrators.
  4. Before integration, estimate call volume and reserve sufficient quota in team quotas to avoid frequently triggering insufficient quota or rate limiting.