export type ModelProvider = 'openai' | 'anthropic' | 'ollama' | 'openrouter' | 'minimax' | 'groq' | 'gemini' | 'perplexity' | 'glm';
export interface ModelConfig {
    provider: ModelProvider;
    model: string;
    baseUrl?: string;
    apiKey?: string;
    costPerMillionInput: number;
    costPerMillionOutput: number;
    maxTokens?: number;
    temperature?: number;
    /** Streaming callback — called with each chunk as it arrives */
    stream?: (chunk: string) => void;
    /** Request JSON-mode response where supported (OpenAI, Anthropic, Groq, OpenRouter) */
    jsonMode?: boolean;
}
export interface ModelabConfig {
    models: Record<string, ModelConfig>;
    budget: {
        maxPerRun: number;
        maxPerExperiment: number;
        trackCosts: boolean;
    };
    evalModel: string;
    parallelism: number;
    cache?: {
        enabled: boolean;
        ttlMs?: number;
    };
    export?: {
        defaultFormat: 'json' | 'md' | 'html';
        outputDir?: string;
    };
}
export interface ResearchGoal {
    id: string;
    question: string;
    goal: string;
    qualityThreshold: number;
    maxIterations: number;
    arms: ExperimentArm[];
}
export interface ExperimentArm {
    id: string;
    name: string;
    /** Mustache-style template: {{question}}, {{goal}}, custom vars */
    promptTemplate: string;
    model: string;
    variables?: Record<string, string>;
    /** Override the model's default temperature for this arm */
    temperature?: number;
    /** Temperature sweep: run this arm at each of these temperatures and compare scores.
     *  When set, `temperature` is ignored and the arm is fanned out into multiple sub-arms.
     *  Example: [0, 0.3, 0.7, 1.0] */
    temperatureSweep?: number[];
}
export interface ExperimentResult {
    armId: string;
    /** The actual model name used (e.g. "claude-sonnet-4-6", "glm-4-flash") */
    model: string;
    /** The model config key used (e.g. "balanced", "fast", "reasoning") — used for routing history lookups */
    modelKey: string;
    output: string;
    /** First 200 chars of output — for quick preview in tables/history */
    outputPreview: string;
    /** True if output was truncated (>200 chars stored) */
    outputTruncated: boolean;
    score: number | null;
    /** If scoring/parsing failed, this is the error message */
    scoreError?: string | null;
    costUsd: number;
    tokensUsed: {
        input: number;
        output: number;
    };
    durationMs: number;
    /** Time-to-first-token in ms — 0 for non-streaming calls */
    latencyMs: number;
    timestamp: string;
    iteration: number;
    /** Whether this result came from cache */
    cached?: boolean;
    /** Run ID — always set when stored in memory */
    runId: string;
    /** Goal ID — always set when stored in memory */
    goalId: string;
}
export interface RunLog {
    goalId: string;
    runId: string;
    status: 'running' | 'completed' | 'quality_reached' | 'budget_exceeded' | 'failed';
    startedAt: string;
    completedAt: string;
    totalCostUsd: number;
    bestResult: ExperimentResult | undefined;
    allResults: ExperimentResult[];
    /** Hash of the question — used for cache lookup */
    questionHash?: string;
}
export interface PersistedResult extends ExperimentResult {
    runId: string;
    goalId: string;
}
export type ExportFormat = 'json' | 'md' | 'html';
export interface ExportOptions {
    format: ExportFormat;
    includeScores?: boolean;
    includeCost?: boolean;
    includeMetadata?: boolean;
    theme?: 'light' | 'dark';
    shareMode?: boolean;
}
export interface PromptTemplate {
    id: string;
    name: string;
    description: string;
    promptTemplate: string;
    recommendedModels?: string[];
    tags?: string[];
}
//# sourceMappingURL=types.d.ts.map