go-llm/docs/api.md

API

Generated from the package comment and every exported declaration. Do not edit. Regenerate with UPDATE_API_DOCS=1 go test -run TestAPIDoc -count=1.

Package gollm

Package gollm is a chat LLM wire client: Chat, Stream, and List Models through an Ollama-native or OpenAI-compatible Adapter.

Types

Adapter

type Adapter string

Adapter selects the wire protocol a Client speaks. AdapterOpenAI uses POST /v1/chat/completions and GET /v1/models. AdapterOllama uses POST /api/chat and GET /api/tags.

AdapterOpenAI

AdapterOllama

const (
	// AdapterOpenAI selects the OpenAI-compatible protocol.
	AdapterOpenAI	Adapter	= "openai"
	// AdapterOllama selects the Ollama native protocol.
	AdapterOllama	Adapter	= "ollama"
)

Client

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

Client calls one Adapter at one base URL. An empty API key sends no Authorization header. A non-empty key is sent as a Bearer token on every request.

NewClient

func NewClient(adapter Adapter, baseURL, apiKey string) (*Client, error)

NewClient checks adapter and baseURL. It returns an error when adapter is not AdapterOpenAI or AdapterOllama, or when baseURL is empty or only whitespace. A trailing slash on baseURL is removed. apiKey may be empty.

Client.Chat

func (c *Client) Chat(ctx context.Context, messages []Message, opt Options) (string, error)

Chat sends messages and returns the assistant text. It returns an error when ctx is canceled, the HTTP status is not 2xx, the body cannot be decoded, or an OpenAI response contains no choices. An HTTP failure includes a short prefix of the response body.

Client.ListModels

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

ListModels returns model names reported by the server. AdapterOpenAI reads data[].id from GET /v1/models. AdapterOllama reads models[].name from GET /api/tags. It returns an error when ctx is canceled, the HTTP status is not 2xx, or the body cannot be decoded. Empty names are omitted.

Client.Stream

func (c *Client) Stream(ctx context.Context, messages []Message, opt Options, yield func(chunk string) error) error

Stream sends the same request as Chat with streaming enabled and calls yield for each non-empty text chunk, in order. It returns the error from yield. It also returns an error when ctx is canceled, the HTTP status is not 2xx, or a stream line cannot be decoded. AdapterOpenAI ends the stream when a data line is the token DONE. AdapterOllama reads one JSON object per line and yields message content.

Message

type Message struct {
	Role	Role
	Content	string
}

Message is one chat turn. Role and Content are sent unchanged.

Options

type Options struct {
	Model		string
	Temperature	*float64
	MaxTokens	*int
}

Options configures one Chat or Stream call. Model is the server-side model name. The Client does not check that it is set; the server rejects an empty model. A nil Temperature or MaxTokens is left out of the request. On AdapterOllama, MaxTokens is sent as num_predict.

Role

type Role string

Role is the speaker of one Message.

RoleSystem

RoleUser

RoleAssistant

const (
	// RoleSystem is a system instruction.
	RoleSystem	Role	= "system"
	// RoleUser is the caller's turn.
	RoleUser	Role	= "user"
	// RoleAssistant is a previous model turn.
	RoleAssistant	Role	= "assistant"
)