import type { DocumentFileType, IngestOptions } from '../types/index.js';
import { LocalEmbedder } from './embedder.js';
import { DocumentStore } from './store.js';
/**
 * Callback for entity extraction — caller provides their own LLM.
 */
export type EntityExtractionFn = (prompt: string) => Promise<string>;
export interface IngestionResult {
    docId: string;
    filename: string;
    chunks: number;
    tokens: number;
    entities: number;
    relationships: number;
    embeddingsGenerated: boolean;
}
/**
 * Detect file type from extension.
 */
export declare function detectFileType(filePath: string): DocumentFileType | null;
/**
 * Extract text from a file based on its type.
 *
 * For simple formats (md, txt, csv, json, html), reads directly.
 * For complex formats (pdf, docx, xlsx, pptx), the caller should
 * pre-extract text using MCP servers (MarkItDown, pdf-reader, etc.)
 * and pass it to ingestText() instead.
 */
export declare function extractText(filePath: string, fileType: DocumentFileType): string;
/**
 * Full ingestion pipeline for a file on disk.
 *
 * 1. Detect file type
 * 2. Extract text
 * 3. Chunk the text
 * 4. Generate embeddings (if Ollama available)
 * 5. Extract entities (if LLM callback provided)
 * 6. Store everything in SQLite
 */
export declare function ingestFile(filePath: string, store: DocumentStore, options?: IngestOptions & {
    embedder?: LocalEmbedder;
    entityExtractor?: EntityExtractionFn;
}): Promise<IngestionResult>;
/**
 * Ingest pre-extracted text (e.g., from MCP server output).
 * Use this for PDF/DOCX/XLSX that have already been converted to text.
 */
export declare function ingestText(text: string, filename: string, fileType: DocumentFileType, store: DocumentStore, options?: IngestOptions & {
    filepath?: string;
    embedder?: LocalEmbedder;
    entityExtractor?: EntityExtractionFn;
}): Promise<IngestionResult>;
//# sourceMappingURL=ingestion.d.ts.map