model.go

package blades

import (
	"context"
)

// ModelOption configures a single request. Providers may ignore options
// they do not support but should prefer best-effort behavior.
type ModelOption func(*ModelOptions)

// ModelOptions holds common request-time controls.
type ModelOptions struct {
	MaxIterations   int
	MaxOutputTokens int64
	Temperature     float64
	TopP            float64
	ReasoningEffort string
}

// ModelRequest is a multimodal chat-style request to the provider.
type ModelRequest struct {
	Model    string     `json:"model"`
	Tools    []*Tool    `json:"tools,omitempty"`
	Messages []*Message `json:"messages"`
}

// ModelResponse is a single assistant message as a result of generation.
type ModelResponse struct {
	Messages []*Message `json:"message"`
}

// ModelProvider is an interface for multimodal chat-style models.
type ModelProvider interface {
	// Generate Generate executes the request and returns a single assistant response.
	Generate(context.Context, *ModelRequest, ...ModelOption) (*ModelResponse, error)
	// NewStream executes the request and returns a stream of assistant responses.
	NewStream(context.Context, *ModelRequest, ...ModelOption) (Streamer[*ModelResponse], error)
}