import type { Router } from '@cortex/llm';
import { type SQLiteStore, type VectorStore } from '@cortex/graph';
export interface PipelineOptions {
    projectId: string;
    projectName: string;
    projectRoot: string;
    maxFileSize: number;
    batchSize: number;
    projectPrivacyLevel: 'standard' | 'sensitive' | 'restricted';
    mergeConfidenceThreshold: number;
    secretPatterns?: string[];
}
export interface PipelineResult {
    fileId: string;
    entityIds: string[];
    relationshipIds: string[];
    status: 'ingested' | 'failed' | 'skipped';
    error?: string;
}
export declare class IngestionPipeline {
    private router;
    private store;
    private options;
    private vectorStore?;
    private checkedContradictionPairs;
    private compiledSecretPatterns;
    constructor(router: Router, store: SQLiteStore, options: PipelineOptions, vectorStore?: VectorStore);
    /**
     * Scrub secrets from content before sending to cloud LLMs.
     * Only applied for standard privacy (sensitive/restricted use local provider).
     */
    private scrubSecrets;
    ingestFile(filePath: string): Promise<PipelineResult>;
    private extractEntities;
    private inferRelationships;
    /**
     * Generate and store semantic embeddings for the given entities. Best-effort: failures
     * are logged but never fail ingestion. Skipped when no vector store / embedding provider
     * is configured, or for non-standard privacy projects (whose content must not reach the cloud).
     */
    private indexEmbeddings;
    private deduplicateEntities;
}
//# sourceMappingURL=pipeline.d.ts.map