Docs

English

简体中文

English

简体中文

Appearance

Add a Channel ​

Each field on the New Channel form: what it is, where to get it, what it changes.

Open Admin → Channels → Add. The form has these fields.

Required ​

Provider type ​

  • What: which upstream (OpenAI / Anthropic / Azure / Zhipu / DeepSeek / Ollama …).
  • Where: pick from the dropdown.
  • Effect: determines the request protocol (OpenAI-compatible / Anthropic Messages / Azure custom). Wrong choice = nothing works.

Channel name ​

  • What: a free-text label.
  • Where: anything you like (e.g. OpenAI Official, Azure East, Zhipu Production).
  • Effect: display only, no protocol impact.

Base URL ​

  • What: the upstream provider's endpoint URL.
  • Where:
    • OpenAI: https://api.openai.com/v1
    • Azure OpenAI: from your deployment page, e.g. https://<resource>.openai.azure.com/openai/deployments/<dep>
    • Third-party proxy: the URL your vendor gives you
  • Effect: wrong value = 404. Note whether /v1 is required.

API key / credential ​

  • What: the upstream access token.
  • Where: upstream console → API Keys / Access Tokens.
  • Effect: wrong value = 401.
  • Tip: copy whole, no quotes / spaces / newlines.

Models ​

  • What: which models this channel serves (multi-select).
  • Where: options come from Model Price entries you have enabled.
  • Effect: unselected models are not routed here.
  • Tip: set the Model Price first, otherwise calls succeed but no quota is deducted.

Common ​

Weight / Priority ​

  • What: a number; higher = more likely to be picked.
  • Suggestion: cheap-and-fast channels high, expensive-and-slow low.

Concurrency cap ​

  • What: max simultaneous in-flight calls.
  • Effect: exceeding it trips the breaker / re-routes.
  • Suggestion: match your upstream provider's RPM / TPM limits.

Enabled ​

  • What: a toggle.
  • Effect: disabled = router skips.

Advanced (rarely needed) ​

Custom request headers ​

Only when the upstream requires a specific header (e.g. legacy Azure with api-key).

Response JSONPath ​

For non-standard upstream responses; leave blank 99% of the time.

Retry / timeout ​

Bump only if the upstream is slow or flaky.

After save ​

  1. Save.
  2. Open the channel detail.
  3. Channel Test — enter a model name; confirm success.
  4. Once passing, enable the channel.