provider

Documentation

Overview

Package provider defines Alaska's LLM provider abstraction. Concrete providers (Anthropic, OpenAI, OpenAI-compatible) live in subpackages.

Index

• type ChatRequest
• type Message
• type ModelInfo
  • func KnownModels(providerType string) []ModelInfo
• type Provider
• type Role
• type StreamChunk

Types

type ChatRequest

type ChatRequest struct {
	Model       string
	System      string
	Messages    []Message
	MaxTokens   int
	Temperature float64
	// ReasoningEffort maps to OpenAI's `reasoning_effort` ("minimal"/"low"/
	// "medium"/"high") for o-series, gpt-5 and proxies that honour it.
	// Empty string means "don't send the field".
	ReasoningEffort string
}

type Message

type Message struct {
	Role    Role
	Content string
}

type ModelInfo

type ModelInfo struct {
	ID          string
	DisplayName string
	Note        string
}

ModelCatalog is a curated list of models known to work with each provider type. Used by `/model` (interactive picker) and `alaska provider models`. Custom providers of type "openai" with a base_url get no default catalog — users list models via `alaska provider models <name>` which hits the remote /models endpoint.

func KnownModels

func KnownModels(providerType string) []ModelInfo

KnownModels returns a static catalog for a provider type. Empty slice means "no known catalog — discover at runtime or let the user type the id".

type Provider

type Provider interface {
	Name() string
	Stream(ctx context.Context, req ChatRequest) (<-chan StreamChunk, error)
}

type Role

type Role string

const (
	RoleUser      Role = "user"
	RoleAssistant Role = "assistant"
)

type StreamChunk

type StreamChunk struct {
	Delta string
	Err   error
}

StreamChunk carries one piece of the provider's streaming response. Exactly one of Delta or Err is populated per chunk.