Duckle docs
Duckle is the open, single-engine ETL / ELT stack. Write a pipeline in Python, wire it from connectors, or draw it on a canvas, run it at native speed on DuckDB, and ship it as plain files. This is everything you need to go from download to a scheduled production pipeline.
Explore
Getting started
Download, install the DuckDB engine on first launch, and build + run your first source → transform → sink pipeline.
Component reference
All 363 components: 111 sources, 130 transforms, 68 sinks, 28 quality validators, 19 control-flow nodes, 7 code runners.
Integrations directory
Every connector by category - files, databases, warehouses, object stores, streaming, SaaS, NoSQL and vector DBs.
Duckie AI
The on-device assistant, the six AI transforms, and how to clean data before it reaches your models.
Scheduling, MCP & deploy
Cron / interval / file-watch schedules, the MCP server for Claude and other LLMs, connections, contexts, and headless server deployment.
Learn hub
Concepts and guides: ETL vs ELT, change data capture, local-first architecture, RAG, and migration from other tools.
What is Duckle?
A visual data pipeline studio that runs on your laptop. Drag sources, transforms, validators and sinks onto a canvas, wire them, and press Run. Duckle compiles the graph to SQL and executes it through DuckDB - a real columnar engine - with live previews, generated SQL on every node, and zero hidden state.
- An AI assistant in the box. Describe a pipeline in English; Duckie writes the JSON and drops it on the canvas. The model runs locally - no API key, no telemetry.
- 345 components ready to run at install time. Files, lakehouses, SQL databases, warehouses, NoSQL, vector DBs, streaming brokers, SaaS APIs, FTP and IMAP - working today.
- Dives and dashboards. Turn a pipeline result into a local-first, live-querying view and a multi-chart dashboard; ask a plain-language question and AI generates the chart; export a self-contained HTML file or open a standalone share page.
- Run visibility you can trust. Column lineage per node traces each output column back to its source columns, and a failed stage reports the exact compiled SQL and the DuckDB message in the Run view and the NDJSON log.
- Safer loads. Database sinks validate before insert and route rows that do not match the declared types to a dead-letter file, so only clean rows reach the table.
- A self-contained binary you can audit. ~65 MB download; engines install on first launch; workspaces are plain files in a folder you choose.
Built on DuckDB
Every pipeline runs through DuckDB - vectorized, columnar, local. Duckle reads the wider DuckDB family natively: DuckLake tables, MotherDuck, the Quack remote protocol, and the community extensions (httpfs, postgres, iceberg, delta, vss, fts, and more). Duckle is an independent, open-source project by SlothFlowLabs and is not part of, affiliated with, or endorsed by DuckDB Labs or MotherDuck.