01
Workspace
Turn any folder into a review workspace.
Point Retriever at a folder of PDFs, DOCX, emails, PST/MBOX archives, Slack exports, or processed productions. Files stay where they are; state lives under .retriever/ next to them.
Open Source
Document Intelligence
Claude Plugin
Retriever is an open source document intelligence plugin for Claude Code. It turns local folders, productions, mailbox exports, and other supported document collections into a workspace you can search, browse, review, enrich, analyze, and export with Claude.
— Quick start
Up and running in four steps.
Open Claude Code and paste the commands below. Claude does the rest — no terminal, no config files.
01
Download Retriever
Open Claude Code and paste this. It clones the skill into ~/.claude/skills/retriever.
!git clone --single-branch --depth 1 https://github.com/sdemyanov/retriever.git ~/.claude/skills/retriever
02
Install Retriever
Paste this next. Claude runs the setup script and you're ready to go.
› !cd ~/.claude/skills/retriever && ./setup
03
Ingest documents
In a new chat, tell Claude where your files live.
› Ingest this folder /path/to/your/documents
04
Just ask
Talk to it like an assistant — no commands or syntax to learn.
› Show me document distribution by type
— Demo
Watch a real review session, end to end.
From a raw folder to a citation-grounded export, in a single uninterrupted run.
— Try it yourself
Kick the tires on a
real public dataset.
Want to repeat it yourself? Here's the data. Download both the plugin and the dataset, point Retriever at it, and try the exact session from the demo.
- 01
Install Retriever
Follow the two-step install above in Claude Code.
- 02
Clone the dataset using Claude Code
!git clone https://github.com/sdemyanov/retriever-data-public.git
- 03
Ask away
› Ingest ./retriever-data-public and describe what is in there
— Features
What can you do with it?
Six primitives that compose into a real review workflow.
02
Natural language
Search and ask in plain English with grounded answers.
Ask in plain English — no query syntax. Retriever blends keyword and metadata search, and pins every answer to the exact file, page, and Bates number so you can verify the source.
03
AI enrichment
Unlock text in scans, images, and foreign files.
Run OCR on scanned pages, translate foreign-language documents, and auto-generate image descriptions so you can search over them — all inside your workspace.
04
Analytics
Ask for charts and stats.
Request summaries, distributions, timelines, and cross-field breakdowns — Retriever builds them on the fly from your collection. Ask follow-ups, filter, and explore patterns without touching a spreadsheet.
05
Tag & annotate
Tag, label, and enrich documents your way.
Add custom fields, review with versioned custom prompts, generate summaries, extract structured information, or bulk label selected documents as you need.
06
Export
Hand off a real review set.
Export CSVs for QC, HTML preview bundles for sharing, or zip archives with sources, previews, and a portable workspace subset — all scoped to the selection.
— Open source
Bend it to your workflow.
Right from the same terminal.
Retriever is open source and runs inside Claude Code, so customization has no ceiling. Ask Claude to add a new extractor, tweak ranking, change the preview layout, wire in a custom export, or reshape any workflow — in the same terminal you already use to query your documents. Your repo, your rules, your features.