# DocETL

> Processes document collections with LLMs in steps that optimize themselves

Steal this: An LLM pipeline of map, reduce and filter steps that rewrites itself to run cheaper.

- Category: Agents & dev (`agents`)
- Tags: `open-source`, `claude-code`
- Official: https://docetl.org
- GitHub: https://github.com/ucbepic/docetl
- Wunderlust: https://wunderlust.xyz/r/docetl
- Image: https://wunderlust.xyz/media/docetl-poster.jpg
- Video preview: https://wunderlust.xyz/media/docetl.webm
- Added: 2026-11-10

## What it is

An open source system for processing large collections of data with LLMs. You write each operation in plain language (map, reduce, filter) and it runs them in parallel, optimizes the pipeline and returns tables.

## What it’s for

For analyzing hundreds of documents with AI without wiring and tuning every call by hand.

## Benefits

- Operations written in plain language
- Tunes models and prompts for cost and accuracy
- A Claude Code skill helps write the pipeline

## Caveats

- Needs an API key from an LLM provider
- MIT license

## Similar on Wunderlust

- [Unstract](https://wunderlust.xyz/r/unstract.md): No-code platform that pulls data out of documents with LLMs
- [Postman](https://wunderlust.xyz/r/postman.md): API platform to explore, test and document, with skills for agents
- [Open Design](https://wunderlust.xyz/r/open-design.md): Local-first, open source design workspace run by AI agents
- [AI Job Search](https://wunderlust.xyz/r/ai-job-search.md): An open source framework that uses Claude Code for job search and applications
- [LitServe](https://wunderlust.xyz/r/litserve.md): Python framework for serving AI models as APIs with batching and scaling
- [Agency Agents](https://wunderlust.xyz/r/agency-agents.md): Over 230 specialist agents, ready to install across many tools