dlt (data load tool) is an open-source Python library for building data pipelines as code, now extended by dltHub, an agentic data layer that lets coding agents build and run pipelines on any warehouse.
GUI-based ETL tools break down exactly where data engineering gets hard: the obscure internal API, the legacy database, the source no vendor bothered to build a connector for. dlt takes the opposite bet. It is a lightweight Python library — pip install dlt, no backend required — that standardizes the tedious parts of ingestion: schema inference and evolution, incremental loading state, data normalization, and loading into warehouses like Snowflake, BigQuery, Databricks or DuckDB. You keep the full flexibility of plain Python while the library handles the plumbing that makes hand-written scripts fragile.
dltHub, the managed platform on top, is built for the agent era. It ships context pages for 10,000+ sources — auth schemes, pagination, incremental cursors and rate limits already researched — so a coding agent in Claude Code, Cursor or Codex can write a working pipeline for Oracle, SAP HANA or IBM Db2 instead of discovering facts by trial and error. Expectations and schema contracts validate data on the way in, failures reach their owner with a diagnosis attached, and everything runs on managed infrastructure pointed at your warehouse, with lineage and audit trails that regulated industries (GxP, SOX, SR 11-7) can actually defend.
The dlt library is Apache 2.0 licensed and always free for commercial use, with no backend and no usage caps. dltHub, the managed platform with runtime, observability and governed collaboration, is a paid product offered with a trial — pricing scales with team size and pipeline volume. Many teams start and stay on the free library, adopting dltHub when pipelines become business-critical.
Data engineers and Python developers who need custom connectors that GUI tools will never ship, and teams letting AI agents build pipelines under real governance. Non-technical users who want a purely visual, no-code integration tool should look elsewhere.