/** A deterministic, model-neutral token estimate unit. */
export interface EstimatedTokenUnit {
    value: string;
    offset: number;
}
/**
 * Estimate tokens without claiming compatibility with a model tokenizer.
 *
 * Latin-like words, code identifiers, URLs, and repository paths are lexical
 * units. Consecutive Han, Hiragana, Katakana, and Hangul text is divided into
 * two-code-point units so unspaced Japanese is neither one token nor one token
 * per character. Other punctuation is grouped in runs of up to three code
 * points. The algorithm is Unicode-aware, dependency-free, and deterministic,
 * but is only suitable for stable repository thresholds and comparisons.
 */
export declare function estimateTokens(value: string): number;
/** Return the deterministic units used by estimateTokens. */
export declare function estimatedTokenUnits(value: string): EstimatedTokenUnit[];
/** Return Markdown after a complete leading YAML frontmatter block. */
export declare function markdownBody(value: string): string;