Free & Open-Source Data Extraction Tools (2026)

Every list of free data extraction tools hides the same trick: half the entries are 14-day trials, and most of the rest meter you at a few hundred records a month. I checked licenses and free tiers for this list instead of copying vendor marketing pages. Seven entries below plus two honorable mentions, all free to keep using, with the catches written next to them.

Tool	Extracts	License	Free in what way	The catch
Trafilatura	Main content from web pages	Apache-2.0	FOSS, runs local	Drops tables; no JS rendering
Tabula	Tables from text-layer PDFs	MIT	FOSS, offline desktop	Needs Java; scans give it nothing
Docling	Whole documents to Markdown/JSON	MIT	FOSS	Heavy model downloads; slow on CPU
Apache Tika	Text and metadata, 1,000+ formats	Apache-2.0	FOSS	Flat text, no structure
Tesseract	Text out of images (OCR)	Apache-2.0	FOSS	Needs preprocessing; weak on handwriting
WebScraper.io	Point-and-click site scrapes	Proprietary, $0 extension	The extension costs nothing	Hand-built selectors; cloud is paid
Octoparse	Point-and-click, cloud	Freemium	10 tasks, local runs	Paid from $69/month

Ordering logic: how much structure each tool finishes for you. Trafilatura and Tabula hand you shaped output. Tika hands you ingredients. The GUI tools come last because their free tiers are shaped by a sales team, and I’d rather you meet the honest options first.

What I counted as free

Three tests, applied today. First, the license file: I read each one on the project’s repo, and the star counts and commit dates below come from the GitHub API, pulled in October 2026. Second, runnable here: Trafilatura and pdftotext ran on this machine for this article; the rest use commands from their own documentation. Third, free means usable next month at zero cost. Trials are excluded on principle. Where a free plan exists, its caps are stated as published, not rounded in the vendor’s favor.

If the vocabulary around extraction pipelines is new to you, the data extraction guide covers the process end to end. This page is only about tools. One scope line while we’re being honest: “data extraction” also names the ETL world, where databases pour into warehouses. Those tools deserve their own comparison; this list holds to the document and web senses, which is where the search demand actually is.

1. Trafilatura: web pages to clean text

Trafilatura takes a URL, strips navigation, ads, and boilerplate, and returns the article text with whatever metadata the page publishes. Python library and CLI in one package, Apache-2.0, 6,974 stars, actively maintained (version 2.3.1, pushed October 2026).

pip install trafilatura
trafilatura -u "https://en.wikipedia.org/wiki/Data_extraction" --markdown

That exact command, run for this article, returns 419 words of clean Markdown. Abridged:

# Data extraction

**Data extraction** is the act or process of retrieving data out of
(usually unstructured or poorly structured) data sources for further
data processing or data storage (data migration).

Pros

• One dependency, no account, no network calls beyond the page itself
• Output as txt, Markdown, CSV, JSON, XML, or TEI
• --precision and --recall flags trade noise against completeness when the default boilerplate strip is too aggressive

Cons

• Tables vanish. The Wikipedia page’s tables did not survive the run above; if your data is the table, look further down this list
• It fetches raw HTML only, so pages that assemble content with JavaScript come back with holes
• One document at a time. Site-wide crawling is a different job for different tools

2. Tabula: PDF tables to CSV, offline

Tabula is a small desktop app for one job: getting tables out of PDFs that have a text layer. MIT license, 7,489 stars, and the workflow is visual. You open the PDF, draw a box around the table, preview what it caught, and export CSV or JSON. Nothing uploads; the app is fully offline.

The command-line version does the same without the GUI:

java -jar tabula.jar -p all -o tables.csv report.pdf

Pros

• Drawing a box beats fighting page coordinates in code
• A guess mode detects table regions without a drawn area
• Runs offline, so confidential PDFs stay confidential

Cons

• Requires a text layer. Scanned PDFs return nothing, which is what entry 5 is for
• Needs a Java runtime installed
• Development has gone quiet: the repo’s last push was March 2025. It still works on most text-layer PDFs, but budget an alternative (pdfplumber, below) for when one breaks

3. Docling: whole documents to Markdown for LLMs

IBM open-sourced Docling in 2024 and it became the default answer to “how do I feed PDFs and Office files to an LLM”: MIT license, 68,980 stars as of October 2026, the fastest-growing repo on this page. It reads PDF, DOCX, PPTX, XLSX, HTML, and images, runs layout and table-structure models to reconstruct headings and tables, and exports Markdown or JSON.

from docling.document_converter import DocumentConverter

converter = DocumentConverter()
result = converter.convert("report.pdf")
print(result.document.export_to_markdown())

Pros

• Table recognition is the reason to pick it. Merged cells and multi-page tables come out as structures, not soup
• OCR is built in as an option, so scanned pages can pass through the same pipeline
• Everything runs locally. Documents stay on your machine

Cons

• The first run downloads model weights, hundreds of megabytes before the first conversion
• A long PDF on CPU takes minutes, not seconds
• Dependencies pull in torch; the install is not light

The document-versus-web split changes more than the tool choice, and the document data extraction page covers what actually differs when the source is a file instead of a site.

4. Apache Tika: the thousand-format shredder

Tika detects and extracts text and metadata from over a thousand file formats, including legacy .doc and .ppt, emails, and archives. Apache-2.0, 4,098 stars, an Apache Software Foundation project, which means boring, decade-deep stability. The convenient form is a REST server:

docker run -d -p 9998:9998 apache/tika
curl -T report.docx http://localhost:9998/tika > extracted.txt

Pros

• Format coverage nothing else here matches; it also identifies files by content, so a mislabeled .doc that is really RTF gets read anyway
• One API for a whole mixed archive
• Metadata extraction is precise enough to drive a routing pipeline

Cons

• Output is flat text or XHTML. Structure is your problem; Tika tells you what is in the file, not what it means
• The JVM has a memory appetite on large files
• Its optional OCR pass is weaker than a dedicated OCR stage

5. Tesseract: text out of scans

Tesseract is the OCR engine most other tools borrow, including the optional OCR backends in Docling and Tika. Apache-2.0, 77,026 stars, language packs for over 100 languages.

sudo apt install tesseract-ocr
tesseract scan.png out

An out.txt appears next to the image. On a clean, flat, 300-DPI scan of printed text, the result is genuinely good. On a photo of a receipt taken at an angle, it is not, and no flag fixes that.

Pros

• Free, offline, scriptable from any pipeline
• Language packs via -l (tesseract scan.png out -l deu)
• The huge community means every preprocessing question has a written answer somewhere

Cons

• No preprocessing built in. Skew, low contrast, and noise destroy accuracy; you deskew and sharpen with ImageMagick or OpenCV first
• No layout logic. Multi-column pages interleave unless you split the image yourself
• Handwriting remains a weakness

6. WebScraper.io: the free web extractor that stays free

WebScraper is a browser extension for Chrome, Firefox, and Edge. You build a sitemap of selectors (titles, prices, links), add paginators for next-page buttons, and scrape from your own tab. The export is CSV. It is the only free web extractor on this page with no caps at all, because your browser does the work.

Pros

• Point-and-click selection, no code, no account for the extension itself
• Paginators and sub-selectors handle multi-page listings
• Runs inside your logged-in session, so pages behind a login you own are reachable

Cons

• Selectors are hand-built and break silently on a site redesign: you get an empty CSV, not an error
• Long crawls tie up the browser; the tab has to stay open
• Scheduling, proxies, and the API are the paid product, at $50 to $200 per month in the cloud plans

7. Octoparse: free with a meter

Octoparse is a point-and-click extractor with cloud runs and templates for common site types. The free plan, as published on its pricing page: 10 tasks, 1 device, 1 user, local extraction, 2 concurrent runs, and history of your last 5 runs. Paid plans start at $69 per month.

Pros

• Templates get a non-coder to rows of data in minutes
• Local extraction runs on the free tier, so small jobs never touch the meter
• Task configs are reusable when a site’s structure holds

Cons

• Ten tasks fill up fast; every new target means deleting an old one
• Five runs of history makes debugging yesterday’s failed job guesswork
• The features that make cloud extraction convenient, like scheduling and large exports, sit behind the paid plans

Honorable mentions

pdftotext (poppler, GPL) is already installed on most Linux boxes; this one runs version 24.02.0. pdftotext -layout report.pdf - prints the text with columns roughly preserved. It solves the simple case in zero setup time.

pdfplumber (MIT, 10,821 stars) is the Python library for PDF words with coordinates, which is what you want when geometry matters: I reach for it when Tabula’s guess mode misses a table.

Data extraction software free of the trial trap

“Free” means four different things on vendor pages, and the mix explains most bad purchases.

1. FOSS license. Apache, MIT, and BSD code you can run at work, modify, and embed in something you sell. Trafilatura, Tabula, Docling, Tika, Tesseract, and pdfplumber sit here. AGPL projects (Magpie among them) add a share-source obligation if you modify them and serve the result over a network.
2. Free-forever tier. The tool is a paid product, but a capped tier never expires. Octoparse’s free plan lives here, and its caps are the product. WebScraper’s extension sits nearby but uncapped, because your browser is the infrastructure.
3. Free trial. Expires. Most “free data extraction software” results that name document tools are this category in disguise, which is why none appear above.
4. Free for one seat. One device, one user. Fine for a personal project, a trap for a team.

Checking any tool not on this list takes thirty seconds: open the repo, read the LICENSE file, look at the last commit date (a stalled extractor is a broken extractor in six months), and compute the caps against your real volume. That last step is where free plans fail; 10 tasks sounds generous until your tenth target.

Free is the wrong call when volume is high, sites render everything with JavaScript, or a deadline exists. That is services and API territory, and the extraction API comparison walks the cost math.

Where a scraper fits in this list

Everything above extracts from files or single pages. If the source is a live website you need to crawl, log into, or render with JavaScript, you are shopping in the scraper aisle instead. Full disclosure, since Magpie publishes this blog: the Magpie core CLI is open source under AGPL-3.0-or-later (or a commercial licence) and free to run locally; the desktop app is free for 5 runs, then $49 per month. Neither belongs in the main list, which is the point of a sidebar. The open-source scraper comparison covers that aisle honestly.

Free data extraction tools FAQ

What’s the best free data extraction software?

There is no single best; pick by source shape. Web pages: Trafilatura. PDF tables: Tabula or pdfplumber. Whole documents for RAG: Docling. Scans: Tesseract after preprocessing. The paid options and where they beat this stack are compared in the data extraction tools guide.

Are open source data extraction tools legal to use at work?

Yes, license-wise: Apache-2.0, MIT, and BSD all permit commercial use, modification, and embedding. Data legality is a separate question from tool licensing. Public data is generally fine to collect; login-walled or terms-prohibited data is not, whatever the tool cost.

Which free data extractor handles scanned PDFs?

Tesseract, after preprocessing: deskew, boost contrast, 300 DPI minimum. Nothing in this list reads a scan without an OCR step; Tabula and pdftotext need a real text layer. Mixed PDFs route cleanly: text-layer pages to Tabula, scanned pages to Tesseract.

Do free tools work behind logins?

Technically some do; the WebScraper extension scrapes pages inside your logged-in browser session. Whether that is allowed is a terms-of-service question. Scraping behind login violates the terms of most platforms and gets accounts banned. Your own sites, or sites that permit it, are fine.

When is paying the right call?

Volume and walls. If you need 50,000 records a month from sites with anti-bot protection and JS rendering, extraction APIs and services beat the free stack on hours saved, not just capability. Price one month against your record count first.

Start with your source, not the list

Pick by what you are holding. A web page: entry 1. A text-layer PDF table: entry 2. A document pile bound for a RAG build: entry 3. A scan: entry 5. The ten-second test is the first command in this article:

pip install trafilatura && trafilatura -u "https://example.com/article" --markdown

Clean output means your job is covered free. A JS shell or a missing table means you move one entry down and try again. And when the source is a live site rather than a file, the concepts behind the whole pipeline, source to destination, are in the data extraction guide.