Moonshot

Runtime provider id: moonshot. Served through the reasoning-aware openai-compatible adapter, so Kimi’s reasoning_content streams as thinking — the right provider for reaching Kimi through an OpenAI-compatible proxy (e.g. LiteLLM), where the openai provider’s Chat adapter would drop reasoning.

Provider options

Field	Type	Notes
apiKey	string	Required (or set via environment).
baseURL	string	Optional. Defaults to https://api.moonshot.ai/v1 when omitted.
headers	Record<string, string>	Optional. Extra HTTP headers.
reasoning	object	Optional. See Reasoning.

Reasoning

Setting reasoning.effort enables Kimi thinking (thinking: { type: "enabled" }); omit to leave thinking off. summary/display are not used by Moonshot.

Example configs

{
  "name": "Moonshot",
  "provider": "moonshot",
  "options": {
    "apiKey": "sk-..."
  }
}

{
  "name": "Kimi K2",
  "provider": "Moonshot",
  "options": {
    "model": "kimi-k2-0905-preview"
  },
  "default": true
}

Through an OpenAI-compatible proxy (e.g. LiteLLM), with thinking enabled:

{
  "name": "Kimi via Proxy",
  "provider": "moonshot",
  "options": {
    "apiKey": "sk-...",
    "baseURL": "https://litellm.internal/v1",
    "reasoning": { "effort": "high" }
  }
}

{
  "name": "Kimi K2 (thinking)",
  "provider": "Kimi via Proxy",
  "options": {
    "model": "kimi-k2-thinking"
  }
}