import type { LLMProvider } from '../adapter/llm-provider.js';
import { Classifier } from '../pipeline/classifier.js';
import type { KnowledgeEntry, KnowledgeSource } from '../types/index.js';
import { ChunkStrategy, type ChunkOptions } from './chunk-strategy.js';
import { type DocumentParser, type ParsedSection } from './document-parser.js';
import { type AnalysisArtifact } from './analysis-artifact.js';
export interface DocumentExtractionResult {
    entries: KnowledgeEntry[];
    artifact: AnalysisArtifact;
}
export interface DocumentMetadata {
    path: string;
    title?: string;
}
export type DocumentLLMProvider = LLMProvider;
export interface DocumentExtractorOptions {
    /** Minimum content length to consider a section extractable */
    minContentLength?: number;
    /** Minimum confidence threshold (retained for scoring; all entries are 'active') */
    minConfidence?: number;
    /** Custom classifier instance */
    classifier?: Classifier;
    /** Optional LLM provider for semantic extraction */
    llmProvider?: DocumentLLMProvider;
    /** Optional parser override */
    parser?: DocumentParser;
    /** Optional chunk strategy override */
    chunkStrategy?: ChunkStrategy;
    /** Chunk settings used when llmProvider is present */
    chunkOptions?: ChunkOptions;
}
export declare class DocumentExtractor {
    private classifier;
    private minContentLength;
    private minConfidence;
    private llmProvider?;
    private parser;
    private chunkStrategy;
    constructor(options?: DocumentExtractorOptions);
    /**
     * Extract KnowledgeEntry[] from parsed sections.
     * Each section with sufficient content becomes one entry.
     */
    extract(sections: ParsedSection[], source: KnowledgeSource): Promise<KnowledgeEntry[]>;
    extractFromMarkdown(markdown: string, metadata: DocumentMetadata, source: KnowledgeSource, existingEntries?: KnowledgeEntry[]): Promise<KnowledgeEntry[]>;
    extractFromMarkdownWithArtifact(markdown: string, metadata: DocumentMetadata, source: KnowledgeSource, existingEntries?: KnowledgeEntry[]): Promise<DocumentExtractionResult>;
    /**
     * Resolve type: if classifier returns low confidence, use heuristics
     * based on section structure (e.g., methodology for step-like content).
     */
    private resolveType;
    private isValidType;
    private looksLikeMethodology;
    private extractTags;
    private extractDomain;
    private generateTitle;
    private generateSummary;
    /**
     * FR-A02 AC1: Extract from any supported format.
     * Detects format from metadata/content, selects the right parser, and runs extraction.
     * Supports markdown, plain text, HTML, and pre-converted PDF/EPUB content.
     */
    extractFromDocument(content: string, metadata: DocumentMetadata, source: KnowledgeSource, existingEntries?: KnowledgeEntry[], format?: DocumentFormat): Promise<KnowledgeEntry[]>;
    /**
     * FR-A02 AC1+AC5: Extract from any format with Analysis Artifact.
     */
    extractFromDocumentWithArtifact(content: string, metadata: DocumentMetadata, source: KnowledgeSource, existingEntries?: KnowledgeEntry[], format?: DocumentFormat): Promise<DocumentExtractionResult>;
    private dedupeEntries;
    private toKnowledgeEntry;
}
/**
 * Supported document formats for FR-A02 AC1.
 * 'auto' attempts to detect format from metadata or content.
 */
export type DocumentFormat = 'markdown' | 'plaintext' | 'html' | 'auto';
/**
 * Detect document format from file path extension or content heuristics.
 * PDF/EPUB are expected to be pre-converted to text before reaching the extractor.
 */
export declare function detectDocumentFormat(metadata: DocumentMetadata, content: string): DocumentFormat;
export type { ParsedSection } from './document-parser.js';
//# sourceMappingURL=document-extractor.d.ts.map