llm

type APIError ¶

type APIError struct {
	StatusCode int
	Message    string
	Err        error
}

APIError wraps an HTTP error returned by the LLM provider.

func (e *APIError) Error() string

func (e *APIError) Unwrap() error

type ApproxTokenCounter ¶

type ApproxTokenCounter struct{}

ApproxTokenCounter estimates token usage by assuming roughly one token corresponds to four runes.

func (ApproxTokenCounter) Count(msgs ...openai.ChatCompletionMessageParamUnion) int

type ChatCompletionRequest ¶

type ChatCompletionRequest struct {
	Model         string
	Prompt        string
	ContextLength int
	Temperature   *float64
}

type ChatMessage ¶

type ChatMessage = openai.ChatCompletionMessageParamUnion

func TruncateHistory(tc TokenCounter, msgs []ChatMessage, limit int) []ChatMessage

type ChatResponse ¶

type ChatResponse struct {
	Content string // assistant text
	Usage   any
}

ChatResponse is a non-streaming chat response.

type ChatResponseIterator ¶

type ChatResponseIterator iter.Seq2[ChatResponse, error]

ChatResponseIterator is a streaming sequence of chat responses.

type ChatSession ¶

type ChatSession struct {
	// contains filtered or unexported fields
}

ChatSession represents a single conversational context. Not thread safe, create a separate ChatSession per goroutine or protect calls with a mutex.

func NewChat(c *Client, systemPrompt string, opts ...SessionOpt) *ChatSession

func (s *ChatSession) ContextUsed() ContextUsage

func (s *ChatSession) NewChat(opts ...SessionOpt) *ChatSession

func (s *ChatSession) Send(ctx context.Context, req ChatCompletionRequest) (*ChatResponse, error)

func (s *ChatSession) SendStreaming(ctx context.Context, req ChatCompletionRequest) (ChatResponseIterator, error)

type Client ¶

type Client struct {
	// contains filtered or unexported fields
}

Client implements an open ai api compatible client.

func NewClient(opts ...Option) *Client

func (*Client) Close() error

func (c *Client) Embed(ctx context.Context, req EmbedRequest) (*EmbedResponse, error)

func (c *Client) EmbedBatch(ctx context.Context, req EmbedBatchRequest) (*EmbedBatchResponse, error)

func (c *Client) GenerateCompletion(ctx context.Context, req CompletionRequest) (string, error)

func (c *Client) ListModels(ctx context.Context) ([]string, error)

type CompletionRequest ¶

type CompletionRequest struct {
	Model         string
	SystemPrompt  string
	Prompt        string
	ContextLength int
	Temperature   *float64
}

type ContextUsage ¶

type ContextUsage struct{ Used, Max int }

type EmbedBatchRequest ¶

type EmbedBatchRequest struct {
	Model string
	Input []string
}

EmbedBatchRequest contains multiple inputs to embed with a model.

type EmbedBatchResponse ¶

type EmbedBatchResponse struct {
	Vectors [][]float64
	Usage   *openai.CreateEmbeddingResponseUsage
}

type EmbedRequest ¶

type EmbedRequest struct {
	Model string
	Input string
}

EmbedRequest specifies a model and input string for embedding.

type EmbedResponse ¶

type EmbedResponse struct {
	Vector []float64
	Usage  *openai.CreateEmbeddingResponseUsage
}

type Option ¶

type Option func(*config)

Option configures the OpenAI client.

func WithAPIKey(key string) Option

func WithBaseURL(baseURL string) Option

func WithLogger(logger *slog.Logger) Option

func WithModel(model string) Option

func WithTemperature(t *float64) Option

type SessionOpt ¶

type SessionOpt func(*ChatSession)

func WithDefaultContextLength(l int) SessionOpt

func WithSessionLogger(logger *slog.Logger) SessionOpt

func WithSessionTemperature(t *float64) SessionOpt

func WithTokenCounter(tc TokenCounter) SessionOpt

type TokenCounter ¶

type TokenCounter interface {
	Count(msgs ...openai.ChatCompletionMessageParamUnion) int
}

TokenCounter reports the number of tokens in a set of messages.