anthropic

package
v1.0.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Aug 7, 2026 License: MIT Imports: 15 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	RoleUser      = "user"
	RoleAssistant = "assistant"
	RoleSystem    = "system"
)
View Source 
const MaxTokensAnthropicSonnet35 = "max-tokens-3-5-sonnet-2024-07-15" //nolint:gosec // This is not a sensitive value.

MaxTokensAnthropicSonnet35 is the header value for specifying the maximum number of tokens when using the Anthropic Sonnet 3.5 model.

Variables

View Source 
var (
	ErrEmptyResponse            = errors.New("no response")
	ErrMissingToken             = errors.New("missing the Anthropic API key, set it in the ANTHROPIC_API_KEY environment variable")
	ErrUnexpectedResponseLength = errors.New("unexpected length of response")
	ErrInvalidContentType       = errors.New("invalid content type")
	ErrUnsupportedMessageType   = errors.New("unsupported message type")
	ErrUnsupportedContentType   = errors.New("unsupported content type")
)

Functions

func EphemeralCache 

func EphemeralCache() *llms.CacheControl

EphemeralCache creates a standard ephemeral cache control for Anthropic with 5-minute duration.

func EphemeralCacheOneHour 

func EphemeralCacheOneHour() *llms.CacheControl

EphemeralCacheOneHour creates a 1-hour ephemeral cache control for Anthropic.

func MapError 

func MapError(err error) error

MapError maps Anthropic-specific errors to standardized error codes.

func WithBetaHeader 

func WithBetaHeader(header string) llms.CallOption

WithBetaHeader adds a custom beta header for accessing Anthropic's experimental features. This is useful for testing new features before dedicated support is added.

Usage: 

llm.GenerateContent(ctx, messages,
    anthropic.WithBetaHeader("new-feature-2025-01-01"),
)

func WithCacheStrategy 

func WithCacheStrategy(strategy CacheStrategy) llms.CallOption

WithCacheStrategy enables automatic cache control placement based on strategy. This option applies to a single GenerateContent/Call invocation.

Example for AI agent with many tools: 

llm.GenerateContent(ctx, messages,
    anthropic.WithCacheStrategy(anthropic.CacheStrategy{
        CacheTools: true,  // Cache tool definitions
        CacheSystem: true, // Cache system prompt
    }),
)

func WithExtendedOutput 

func WithExtendedOutput() llms.CallOption

WithExtendedOutput enables 128K token output for Claude 3.7+. Standard models are limited to 8K tokens, but this beta feature allows generating much longer responses.

Usage: 

llm.GenerateContent(ctx, messages,
    llms.WithMaxTokens(50000),
    anthropic.WithExtendedOutput(),
)

func WithInterleavedThinking 

func WithInterleavedThinking() llms.CallOption

WithInterleavedThinking enables thinking between tool calls for Claude 3.7+. This allows the model to use reasoning tokens to plan tool usage and interpret results.

Usage: 

llm.GenerateContent(ctx, messages,
    llms.WithTools(tools),
    llms.WithThinkingMode(llms.ThinkingModeMedium),
    anthropic.WithInterleavedThinking(),
)

func WithPromptCaching 

func WithPromptCaching() llms.CallOption

WithPromptCaching enables Anthropic's prompt caching feature. This allows frequently-used prompts and system messages to be cached for improved performance and reduced costs.

Usage: 

llm.GenerateContent(ctx, messages,
    anthropic.WithPromptCaching(),
)

Types

type CacheStrategy 

type CacheStrategy struct {
	// CacheTools enables caching for tool definitions (placed after last tool).
	CacheTools bool
	// CacheSystem enables caching for system messages (placed after system content).
	CacheSystem bool
	// CacheMessages enables caching for conversation history (placed after last message).
	CacheMessages bool
	// TTL specifies cache duration ("5m" or "1h"). Defaults to "5m" if empty.
	TTL string
}

CacheStrategy defines where to apply automatic caching.

type CachedContent 

type CachedContent struct {
	llms.ContentPart
	CacheControl *llms.CacheControl `json:"cache_control,omitempty"`
}

CachedContent represents content with caching instructions for Anthropic. This wraps any ContentPart and adds cache control metadata.

func WithCacheControl 

func WithCacheControl(content llms.ContentPart, control *llms.CacheControl) CachedContent

WithCacheControl wraps content with cache control instructions for Anthropic. This allows explicit control over what content should be cached.

Usage: 

anthropic.WithCacheControl(
    llms.TextPart("long context..."),
    anthropic.EphemeralCache(),
)

type ErrModelRefusal 

type ErrModelRefusal struct {
	Message      string
	Category     string
	Explanation  string
	InputTokens  int
	OutputTokens int
}

ErrModelRefusal is returned when the model declines to respond (Anthropic stop_reason "refusal"), e.g. a creative model such as Claude Fable 5 hitting a content boundary. It is distinct from an empty or failed response; any refusal text the model returned is in Message. Category and Explanation carry the API's stop_details so callers can pick a fallback by classifier; InputTokens and OutputTokens preserve billed usage (a refusal is still billed for what ran).

func (*ErrModelRefusal) Error 

func (e *ErrModelRefusal) Error() string

type LLM 

type LLM struct {
	CallbacksHandler callbacks.Handler
	// contains filtered or unexported fields
}

func New 

func New(opts ...Option) (*LLM, error)

New returns a new Anthropic LLM.

func (*LLM) Call 

func (o *LLM) Call(ctx context.Context, prompt string, options ...llms.CallOption) (string, error)

Call requests a completion for the given prompt.

func (*LLM) GenerateContent 

func (o *LLM) GenerateContent(ctx context.Context, messages []llms.MessageContent, options ...llms.CallOption) (resp *llms.ContentResponse, err error)

GenerateContent implements the Model interface.

type Option 

type Option func(*options)

func WithAnthropicBetaHeader 

func WithAnthropicBetaHeader(value string) Option

WithAnthropicBetaHeader adds the Anthropic Beta header to support extended options.

func WithBaseURL 

func WithBaseURL(baseURL string) Option

WithBaseUrl passes the Anthropic base URL to the client. If not set, the default base URL is used.

func WithDefaultCacheStrategy 

func WithDefaultCacheStrategy(strategy CacheStrategy) Option

WithDefaultCacheStrategy sets the default caching strategy for all requests made by this client. This strategy will be applied to all GenerateContent/Call invocations unless overridden by a call-level WithCacheStrategy option.

This is safe and cost-effective when: - You have stable tools that don't change (CacheTools: true saves 90% on tools from 2nd request) - You use the same system prompt across conversations (CacheSystem: true) - You want automatic conversation history caching (CacheMessages: true for multi-turn)

Example for AI agent with stable tools: 

llm, err := anthropic.New(
    anthropic.WithDefaultCacheStrategy(anthropic.CacheStrategy{
        CacheTools: true,  // Tools cached once, reused forever
        CacheSystem: true, // System prompt cached once
    }),
)

Call-level strategies override client-level on a per-field basis.

func WithHTTPClient 

func WithHTTPClient(client anthropicclient.Doer) Option

WithHTTPClient allows setting a custom HTTP client. If not set, the default value is http.DefaultClient.

func WithLegacyTextCompletionsAPI 

func WithLegacyTextCompletionsAPI() Option

WithLegacyTextCompletionsAPI enables the use of the legacy text completions API.

func WithModel 

func WithModel(model string) Option

WithModel passes the Anthropic model to the client.

func WithToken 

func WithToken(token string) Option

WithToken passes the Anthropic API token to the client. If not set, the token is read from the ANTHROPIC_API_KEY environment variable.

type ToolResult 

type ToolResult struct {
	Type      string `json:"type"`
	ToolUseID string `json:"tool_use_id"`
	Content   string `json:"content"`
}

Source Files

View all Source files

Directories

Path Synopsis
internal
anthropicclient

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.