Documentation
¶
Overview ¶
Package api handles LLM API communication
Package api defines data structures for API communication
Index ¶
- Constants
- func WithPerAttemptTimeout(ctx context.Context, d time.Duration) context.Context
- func WithRetryNotifier(ctx context.Context, fn func(attempt int, err error)) context.Context
- type Client
- func (c *Client) GetModel() string
- func (c *Client) SendMessage(ctx context.Context, prompt string) (string, error)
- func (c *Client) SendMessageWithHistory(ctx context.Context, messages []Message) (string, error)
- func (c *Client) SendMessageWithTools(ctx context.Context, messages []Message, tools []Tool, executor ToolExecutor, ...) (string, []Message, Usage, error)
- func (c *Client) SendMessageWithToolsStreaming(ctx context.Context, messages []Message, tools []Tool, executor ToolExecutor, ...) (string, []Message, Usage, error)
- func (c *Client) SetModel(model string)
- type ClientOption
- type ContentBlock
- type InputSchema
- type Message
- type MultiExecutor
- type PropertyDef
- type StatusError
- type Tool
- type ToolCallResult
- type ToolExecutor
- type Usage
Constants ¶
const ( DefaultMaxTokens = 8192 MessagesEndpoint = "/v1/messages" AnthropicVersion = "2023-06-01" )
Constants for API configuration
Variables ¶
This section is empty.
Functions ¶
func WithPerAttemptTimeout ¶
WithPerAttemptTimeout returns a context that will apply the given timeout to each individual request attempt. This allows retrying after a timeout, as each retry gets a fresh timeout rather than sharing an already-expired one.
Types ¶
type Client ¶
type Client struct {
// contains filtered or unexported fields
}
Client handles communication with the LLM API
func NewClient ¶
func NewClient(baseURL, apiKey string, opts ...ClientOption) *Client
NewClient creates a new API client with the given base URL and API key
func (*Client) SendMessage ¶
SendMessage sends a single user message and returns the response text.
func (*Client) SendMessageWithHistory ¶
SendMessageWithHistory sends a conversation history and returns the response text.
func (*Client) SendMessageWithTools ¶
func (c *Client) SendMessageWithTools(ctx context.Context, messages []Message, tools []Tool, executor ToolExecutor, onCall ...func(ToolCallResult)) (string, []Message, Usage, error)
SendMessageWithTools sends a conversation with tool definitions, executing any tool calls the model makes and continuing the loop until the model returns a final text response. It returns the final text response, the full accumulated message history (including tool_use/tool_result exchanges and the final assistant message), the total token usage across all API calls in the loop, and any error. The optional onCall callback is invoked after each tool execution with the result.
func (*Client) SendMessageWithToolsStreaming ¶
func (c *Client) SendMessageWithToolsStreaming(ctx context.Context, messages []Message, tools []Tool, executor ToolExecutor, onToken func(string), onCall ...func(ToolCallResult)) (string, []Message, Usage, error)
SendMessageWithToolsStreaming is like SendMessageWithTools but streams text tokens via the SSE API. onToken is called for each text token as it arrives; it may be nil. Tool-calling turns are handled the same way as in SendMessageWithTools.
type ClientOption ¶
type ClientOption func(*Client)
ClientOption is a function that configures a Client
func WithHTTPClient ¶
func WithHTTPClient(httpClient *http.Client) ClientOption
WithHTTPClient sets a custom HTTP client for the API client
func WithMaxRetries ¶
func WithMaxRetries(n int) ClientOption
WithMaxRetries sets the maximum number of retry attempts after a retryable error. Default is 0 (no retries).
func WithMaxTokens ¶
func WithMaxTokens(n int) ClientOption
WithMaxTokens sets the maximum number of tokens in API responses. If not set, DefaultMaxTokens is used.
func WithModel ¶
func WithModel(model string) ClientOption
WithModel sets the model to use for API requests
func WithRetryDelay ¶
func WithRetryDelay(d time.Duration) ClientOption
WithRetryDelay sets the base delay between retry attempts. The actual delay doubles with each attempt (exponential backoff). Default is 0 (no delay).
func WithSystemPrompt ¶
func WithSystemPrompt(s string) ClientOption
WithSystemPrompt sets the system prompt sent with every request.
type ContentBlock ¶
type ContentBlock struct {
Type string `json:"type"`
// text blocks
Text string `json:"text,omitempty"`
// tool_use blocks (assistant → us)
ID string `json:"id,omitempty"`
Name string `json:"name,omitempty"`
Input map[string]any `json:"input,omitempty"`
// tool_result blocks (us → assistant)
ToolUseID string `json:"tool_use_id,omitempty"`
Content string `json:"content,omitempty"`
}
ContentBlock represents a single content block in a message or response. Type is one of "text", "tool_use", or "tool_result".
type InputSchema ¶
type InputSchema struct {
Type string `json:"type"`
Properties map[string]PropertyDef `json:"properties"`
Required []string `json:"required,omitempty"`
}
InputSchema is the JSON Schema for a tool's input object.
type Message ¶
Message represents a single message in the conversation. Content may be a plain string or a []ContentBlock (for tool_use/tool_result).
type MultiExecutor ¶
type MultiExecutor struct {
// contains filtered or unexported fields
}
MultiExecutor dispatches ExecuteTool calls to the registered executor for each tool name.
func NewMultiExecutor ¶
func NewMultiExecutor() *MultiExecutor
NewMultiExecutor returns an empty MultiExecutor.
func (*MultiExecutor) ExecuteTool ¶
ExecuteTool dispatches to the executor registered for name. Returns an error if no executor is registered for name.
func (*MultiExecutor) Register ¶
func (m *MultiExecutor) Register(name string, exec ToolExecutor)
Register associates name with the given executor.
type PropertyDef ¶
PropertyDef describes a single property in an InputSchema.
type StatusError ¶
StatusError is returned when the API responds with a non-200 status code.
func (*StatusError) Error ¶
func (e *StatusError) Error() string
type Tool ¶
type Tool struct {
Name string `json:"name"`
Description string `json:"description"`
InputSchema InputSchema `json:"input_schema"`
}
Tool describes a tool the model may call.
type ToolCallResult ¶
ToolCallResult records a single tool execution that occurred during a request.
type ToolExecutor ¶
ToolExecutor executes a named tool with the given input and returns output.
Source Files