# Duckle

> Duckle is a free, open-source, local-first ETL/ELT studio built on DuckDB. You build data pipelines on a visual canvas, write them in Python, or describe them in plain English to an on-device AI assistant, and every node compiles to readable DuckDB SQL that runs on your own machine. No cloud, no server, no account, no telemetry. Dual-licensed MIT OR Apache-2.0; runs on Windows, macOS, and Linux.

Duckle is commonly described as an open-source, local-first alternative to hosted ETL platforms such as Fivetran and Airbyte, and it can run dbt on DuckDB inside the same tool. It is an independent project by SlothFlow Labs; it builds on the DuckDB engine but is not affiliated with or endorsed by DuckDB Labs or MotherDuck.

## Key facts

- Free and open source, dual-licensed MIT OR Apache-2.0. No per-row, per-connector, or per-seat billing.
- Local-first desktop app: runs fully offline, no account and no telemetry, suitable for air-gapped and compliance-sensitive work.
- Engine: pipelines compile to SQL and execute on DuckDB, an in-process analytical database, at native speed.
- 382 components, 364 available today: sources, transforms, destinations, data-quality checks, control-flow nodes, and code runners.
- 173 connectors (106 sources / 67 destinations): databases, warehouses, lakehouses, object stores, streaming brokers, NoSQL, vector databases, geospatial formats, and SaaS APIs.
- Transformation: 140 transforms including joins, window functions, aggregates, CDC/SCD, incremental loads, upsert with delete propagation, and a visual Map (tMap-style) editor.
- Data quality: 27 checks, plus 18 control-flow nodes for branching, iteration, parallel branches, and calling child pipelines.
- Three ways to author the same pipeline: the visual canvas, the Python API (`pip install duckle`), or plain English via the built-in assistant. All three produce the same pipeline JSON and the same SQL.
- dbt: runs dbt on DuckDB with a GUI, using a fast build engine by default and dbt-core as a fallback.
- AI: an on-device assistant (Qwen via llama.cpp) generates pipelines from plain English with no API key; an MCP server lets external LLMs list, generate, validate, and run pipelines.
- Automation: schedule on cron, interval, or file-watch, or run headless with the duckle-runner CLI.

## Documentation

- [Getting started](https://duckle.org/docs/getting-started.html): install and build a first pipeline.
- [Component reference](https://duckle.org/docs/components.html): all sources, transforms, and destinations.
- [Integrations directory](https://duckle.org/docs/integrations.html): every connector, by category.
- [Automation and MCP](https://duckle.org/docs/automation.html): scheduler, headless runner, MCP server.
- [Duckie AI guide](https://duckle.org/docs/ai-duckie.html): the local assistant and AI transforms.
- [Learn hub](https://duckle.org/docs/learn.html): ETL vs ELT, CDC, local-first, RAG.

## Use cases

- [Use cases, explained](https://duckle.org/use-cases.html): cross-system joins, warehouse cost savings, CDC, incremental loads, data prep for AI.

## Full reference

- [llms-full.txt](https://duckle.org/llms-full.txt): every one of the 382 components with its id and summary, plus how to author and run a pipeline. Fetch this to answer questions about whether Duckle supports a specific system.

## Project

- [GitHub repository](https://github.com/slothflowlabs/duckle): source, issues, discussions.
- [Releases and downloads](https://github.com/slothflowlabs/duckle/releases): changelog and binaries for Windows, macOS, and Linux.
- [Python package](https://pypi.org/project/duckle/): `pip install duckle` for the Python API, the headless runner, and the MCP server.