/** * AnthropicAdapter — native Anthropic Messages API adapter (Issue 001). * * Anthropic does NOT speak the OpenAI /v1/chat/completions protocol, so it * gets a small native adapter instead of the shared OpenAICompatAdapter: * * POST https://api.anthropic.com/v1/messages * headers: x-api-key, anthropic-version: 2023-06-01, content-type * body: { model, max_tokens, system?, messages: [{role, content}] } * * Streaming uses Anthropic's event-stream protocol (content_block_delta → * text_delta). Wire-token usage comes from the final message_delta event * (M2.2 measured cost) and input_tokens/output_tokens on non-stream responses. * * Error mapping keeps the shared classification contract (the message text is * what classifyFallbackError buckets): 401/403→auth, 429→rate-limit, * 5xx→server, fetch→network, abort→timeout. */ import { InferenceProvider, ModelDescriptor } from './interface.js'; import { InferenceOptions, ProviderConfig } from '../config/types.js'; export declare class AnthropicAdapter implements InferenceProvider { readonly name = "Anthropic"; private config; private baseUrl; constructor(config: ProviderConfig); private headers; generate(prompt: string, options?: InferenceOptions): Promise; generateStream(prompt: string, options: InferenceOptions | undefined, onToken: (token: string) => void): Promise; isAvailable(): Promise; getInfo(): string; listModels(): Promise; } //# sourceMappingURL=anthropic-adapter.d.ts.map