Alembic: Data Distiller

关于这款应用

Alembic by Luke Mclaughlin is a professional, private data-preparation workspace designed to turn messy source material into dependable training datasets and retrieval-ready collections. It supports importing CSV, TSV, JSON, JSONL, SQLite, Markdown, HTML, plain text, code, or entire document folders, automatically detecting structure, encoding, delimiters, types, and import problems before processing begins.

Users can design repeatable pipelines composed of deterministic steps including normalization, null handling, type inference, exact and near-duplicate removal, quality filtering, token counting, semantic chunking, privacy scrubbing, decontamination, schema shaping, and dataset splitting. Steps can be disabled, reordered, inspected, and replayed. The tool prepares clean records for Alpaca, OpenAI, Anthropic, ChatML, and DPO structures, building reproducible splits, sentence-aware token-budgeted chunks, heading paths, metadata, real cl100k token counts, and embedding-ready JSONL.

Privacy and quality controls detect and redact emails, phone numbers, addresses, credentials, API keys, financial identifiers, and other sensitive patterns. Users can screen against evaluation sets, identify exact and fuzzy duplicates, and inspect quality metrics before export. The deterministic pipeline works locally by default. Optional provider augmentation uses a configured provider; credentials stay in Keychain and selected content goes directly to that provider.

Importing, the complete deterministic pipeline, preview, dataset reports, and export are free forever. Alembic Pro is an optional auto-renewable subscription with monthly and annual choices and an eligible 7-day free trial. Pro unlocks LLM-assisted Q&A generation, quality scoring, rewriting, classification, and preference-pair creation. The current localized price and trial eligibility are always supplied by the App Store before purchase. Restore Purchases and Manage Subscription are available in Settings.