/** * BenchAgent - Core agent for running benchmarks using Vercel AI SDK * * Supports multiple providers: OpenAI, Anthropic, Google, xAI, Mistral, Groq * Provides LLM interaction with retry logic, token tracking, and cost calculation. */ import { TokenUsage, CostInfo } from './cost-tracker'; import { ProviderName } from './providers'; export interface AgentConfig { name?: string; model?: string; provider?: ProviderName; instructions?: string; maxRetries?: number; timeout?: number; } export interface TestResult { test_name: string; prompt: string; response: string | null; model: string; provider: ProviderName; agent_name: string; execution_time: number; status: 'success' | 'error'; error?: string; retry_attempts?: number; timestamp: string; token_usage?: TokenUsage; cost?: CostInfo; } /** * A multi-provider agent wrapper for running LLM benchmarks. * Supports OpenAI, Anthropic, Google, xAI, Mistral, Groq via Vercel AI SDK. */ export declare class BenchAgent { private name; private modelString; private provider; private instructions; private maxRetries; constructor(config?: AgentConfig); /** * Get the model name. */ getModel(): string; /** * Get the provider name. */ getProvider(): ProviderName; /** * Get the agent name. */ getName(): string; /** * Get full model identifier (provider/model). */ getFullModelId(): string; /** * Run a single benchmark test with retry logic. */ runTest(prompt: string, testName?: string): Promise; /** * Run multiple benchmark tests. */ runMultipleTests(tests: Array<{ prompt: string; name?: string; }>): Promise; private extractTokenUsage; private calculateCost; private createErrorResult; private sleep; } //# sourceMappingURL=agent.d.ts.map