Rate limiting

export declare class ConcurrencyLimiter {
    constructor(maxConcurrent: number, timeoutMs?: number);
    acquire(): Promise<() => void>;
    get pending(): number;
    get active(): number;
    get available(): number;
}

export interface QuotaConfig {
    tenantQuotas?: Record<string, {
        tokensPerDay: number;
        costPerDay: number;
    }>;
    defaultQuota?: {
        tokensPerDay: number;
        costPerDay: number;
    };
    priorityLevels?: Record<string, "p0" | "p1" | "p2">;
}

export interface RateLimitConfig {
    maxTokensPerMinute?: number;
    maxTokensPerHour?: number;
    maxRequestsPerMinute?: number;
    maxConcurrent?: number;
    perTenant?: boolean;
    perUser?: boolean;
    onLimitReached?: "queue" | "reject" | "degrade";
    degradeStrategy?: {
        useCheaperModel: ModelProvider;
        reduceMaxTokens: number;
    };
}

export interface RateLimitScope {
    tenantId?: string;
    userId?: string;
    agentName?: string;
}

export interface RateLimitStatus {
    allowed: boolean;
    remaining: number;
    resetMs: number;
    degraded?: boolean;
}

export declare class TokenRateLimiter {
    constructor(config: RateLimitConfig);
    check(scope: RateLimitScope, tokensToAcquire?: number): RateLimitStatus;
    acquire(estimatedTokens: number, scope: RateLimitScope): RateLimitStatus;
    record(actualTokens: number, estimatedTokens: number, scope: RateLimitScope): void;
    getUsage(scope: RateLimitScope): {
        minuteTokens: number;
        minuteRequests: number;
        hourlyTokens: number;
    };
    reset(scope?: RateLimitScope): void;
}