kimi

Documentation

Overview

Package kimi implements thinking configuration for Kimi (Moonshot AI) models.

Kimi models use the OpenAI-compatible reasoning_effort format for enabled thinking levels, but use thinking.type=disabled when thinking is explicitly turned off.

Index

• type Applier
  • func NewApplier() *Applier
  • func (a *Applier) Apply(body []byte, config thinking.ThinkingConfig, modelInfo *registry.ModelInfo) ([]byte, error)

Types

type Applier

type Applier struct{}

Applier implements thinking.ProviderApplier for Kimi models.

Kimi-specific behavior:

• Enabled thinking: reasoning_effort (string levels)
• Disabled thinking: thinking.type="disabled"
• Supports budget-to-level conversion

func NewApplier

func NewApplier() *Applier

NewApplier creates a new Kimi thinking applier.

func (*Applier) Apply

func (a *Applier) Apply(body []byte, config thinking.ThinkingConfig, modelInfo *registry.ModelInfo) ([]byte, error)

Apply applies thinking configuration to Kimi request body.

Expected output format (enabled):

{
  "reasoning_effort": "high"
}

Expected output format (disabled):

{
  "thinking": {
    "type": "disabled"
  }
}