import { MultiDbManager } from '../core/multi-db-manager';
import { EmbeddingGenerator } from './embedding-generator';
/**
 * Coordinates embedding generation for all 5 dimensions.
 * Syncs embeddings with 5D-DBs using hash-based change detection.
 */
export declare class EmbeddingPipeline {
    private dbManager;
    private embeddingGenerator;
    private moduleApi;
    private symbolApi;
    private dependencyApi;
    private adrApi;
    private changeApi;
    private moduleSummarizer;
    constructor(dbManager: MultiDbManager, embeddingGenerator: EmbeddingGenerator);
    /**
     * Syncs embeddings for all dimensions.
     * Only generates embeddings for changed/new entities.
     */
    syncEmbeddings(pluginId: string): Promise<void>;
    /**
     * Syncs embeddings for a specific dimension.
     */
    syncDimension(dimension: 'X' | 'Y' | 'Z' | 'W' | 'T', pluginId: string, model: string): Promise<void>;
    /**
     * Gets all entities for a dimension.
     */
    private getEntitiesForDimension;
    /**
     * Extracts content for embedding based on dimension.
     */
    private extractContentForEmbedding;
    /**
     * Gets external ID for an entity.
     */
    private getExternalId;
    /**
     * Computes content hash for change detection.
     */
    private computeContentHash;
    /**
     * Estimates token count from content length (Markdown-optimized).
     * Markdown has different token densities:
     * - Code blocks: ~3 chars/token (more tokens per char due to formatting)
     * - Normal text: ~4 chars/token (standard estimate)
     * - Tables: ~2.5 chars/token (many separators increase token count)
     */
    private estimateTokens;
    /**
     * Optimizes module documentation for embedding while preserving semantic meaning.
     * Strategy: Keep structure and important information, remove detailed tables.
     * Aggressively truncates code blocks to signatures only, limits table rows.
     *
     * @param content Full markdown content
     * @param maxTokens Maximum tokens allowed
     * @returns Optimized content
     */
    private optimizeModuleContentForEmbedding;
    /**
     * Extracts only the structure of a module (header, interface/method names, signatures).
     * Used for hierarchical embeddings where details are in Y-Dimension (Symbols).
     * Only keeps: Headers, interface/method/class/variable names, first line of code blocks (signatures).
     * Removes: All tables, code block bodies, normal text.
     *
     * @param content Full markdown content
     * @returns Module structure (header + names + signatures, no details)
     */
    private extractModuleStructure;
}
//# sourceMappingURL=embedding-pipeline.d.ts.map