/**
 * Transcript Ingester
 *
 * Walks ~/.claude/projects/**‌/*.jsonl, parses Claude Code conversation
 * transcripts, and indexes human + assistant prose into Gaia's FTS5 store.
 *
 * Design:
 * - Streams JSONL line-by-line (never loads full file into memory)
 * - Incremental: tracks last-ingested byte offset per file in SQLite
 * - Dedupes on (file_path, message_index) to survive re-runs
 * - Skips noise: tool_result, system messages, sidechain, pure tool_use-only turns
 */
import Database from 'better-sqlite3';
import type { MemoriaStorage } from './storage.js';
export interface IngestOptions {
    /** Root directory to walk. Defaults to ~/.claude/projects */
    rootDir?: string;
    /** Substring to match against project slug (e.g. "foundation") */
    projectFilter?: string;
    /** Only ingest messages newer than this ISO timestamp */
    since?: string;
    /** Walk and report without writing to DB */
    dryRun?: boolean;
}
export interface IngestStats {
    filesFound: number;
    filesSkipped: number;
    messagesIngested: number;
    messagesSkipped: number;
    bytesProcessed: number;
    projectSlugs: string[];
    dateRange: {
        earliest: string | null;
        latest: string | null;
    };
    parseErrors: number;
    parseErrorSamples: string[];
    dryRun: boolean;
}
/** A single parsed message ready for indexing */
export interface ParsedMessage {
    sessionId: string;
    messageIndex: number;
    timestamp: number;
    role: 'user' | 'assistant';
    text: string;
    projectSlug: string;
    cwd: string | null;
    filePath: string;
    lineOffset: number;
}
export declare const INGEST_STATE_DDL = "\n  CREATE TABLE IF NOT EXISTS transcript_ingest_state (\n    file_path TEXT NOT NULL PRIMARY KEY,\n    last_byte_offset INTEGER NOT NULL DEFAULT 0,\n    last_ingested_at INTEGER NOT NULL,\n    messages_ingested INTEGER NOT NULL DEFAULT 0\n  );\n\n  CREATE TABLE IF NOT EXISTS transcript_dedup (\n    file_path TEXT NOT NULL,\n    message_index INTEGER NOT NULL,\n    memory_id TEXT NOT NULL,\n    PRIMARY KEY (file_path, message_index)\n  );\n\n  CREATE INDEX IF NOT EXISTS idx_dedup_file ON transcript_dedup(file_path);\n";
export declare class TranscriptIngester {
    private db;
    private memoria;
    constructor(memoriaDb: Database.Database, memoria: MemoriaStorage);
    private ensureSchema;
    ingest(options?: IngestOptions): Promise<IngestStats>;
    private collectJsonlFiles;
    private processFile;
    private streamFile;
    private updateIngestState;
    private updateDateRange;
    /** Returns ingest state rows for testing incremental behavior */
    getIngestState(): Array<{
        file_path: string;
        last_byte_offset: number;
        messages_ingested: number;
    }>;
    /** Returns dedup row count for a given file */
    getDedupCount(filePath: string): number;
    /** Nano-id helper (re-exported so tests can use it) */
    static generateId(): string;
}
/**
 * MemoriaStorage keeps its `db` private. We expose it through this module
 * by accepting the db instance directly (tests pass an in-memory DB;
 * the MCP tool passes the real db through getInternalDb).
 */
export declare function createIngester(db: Database.Database, memoria: MemoriaStorage): TranscriptIngester;
//# sourceMappingURL=transcript-ingester.d.ts.map