extract

Extracts key facts from text, files, and URLs in configurable output formats.

$ npx arrey@latest add extract

Input Schema

Field	Type
content	string
format?	string
maxItems?	number
instruction?	string

Output Schema

Field	Type
extraction	string
format	string
chunks	number
model	string

Examples

Extract key facts as bullet points

{
  "content": "./meeting-notes.md",
  "format": "bullets"
}

Extract from a URL as JSON

{
  "content": "https://example.com/brief",
  "format": "json"
}

Limit output to top 5 findings

{
  "content": "paste text here",
  "format": "bullets",
  "maxItems": 5,
  "instruction": "Prioritize legal and financial risk facts."
}

README Reference

arrey/tools/extract

Extracts key facts from text, files, and URLs into configurable formats. Handles long documents automatically via chunk extraction and merge. This code is yours - edit it freely.

Usage

CLI

arrey extract ./notes.md
arrey extract ./notes.md --format json
arrey extract https://example.com/brief --format table
arrey extract "paste text here" --format bullets --maxItems 5

SDK

import { arrey } from 'arrey'

const result = await arrey.run('extract', {
  content: './research.md',
  format: 'json'
})

console.log(result.extraction)

Direct import (tool-local helper)

import { extract } from './arrey/tools/extract'

const result = await extract({
  prompt: './research.md',
  format: 'json',
  temp: 0.2,
  model: 'gpt-4.1-mini'
})

console.log(result.extraction)

Agent tool (Vercel AI SDK)

import { arrey } from 'arrey'
import { extract } from './arrey/tools/extract'

const tools = await arrey.toVercelAIFrom([extract])

Formats

Format	Output
`bullets`	5-10 concise factual bullets
`json`	JSON array of facts with confidence labels
`table`	Markdown table (`Fact
`custom`	Whatever you define in `prompt.ts -> formats.custom`

Customization

Tune extraction behavior

Edit prompt.ts -> chunk and combine.

Change output style

Edit prompt.ts -> formats.

Reduce output length

Pass maxItems at runtime:

arrey extract ./contract.md --format bullets --maxItems 3

Tool-specific model override

Set in arrey.config.yaml:

tools:
  extract:
    provider:
      model: gpt-4o

Files

File	Purpose	Edit frequency
`prompt.ts`	Prompts and output formats	Often
`index.ts`	Execution, chunking, merge behavior	Sometimes
`manifest.json`	Metadata and I/O schema	Rarely
`README.md`	This file	Optional