Sygal: Local PDF, Scan and Web-to-Markdown Converter (Show HN)

Sygal is a desktop document-to-Markdown converter from independent Paris-based developer Cyrill Semah that runs conversion on-device and sells for a one-time €25 launch price, but its "100% local" positioning drew pushback on Hacker News because OCR of images and scans is sent to third-party cloud providers using the user's own API key.

The Show HN post reached 31 points and 8 comments. The most substantive criticism targeted the gap between the marketing headline and the actual data flow.

What Sygal does

Sygal converts documents into clean Markdown and structured JSON for use in prompts, knowledge bases, RAG pipelines and agents. It supports 29 formats, including PDF, Word, Excel, PowerPoint, HTML, email and EPUB, plus images, scans, web pages and YouTube links.

The app runs on macOS 12+ and Windows 10+. The vendor states a conversion time of roughly 1.7 seconds per file.

Core features listed on the product page:

  • Markdown output and structured JSON output
  • Native preview of the original document next to the converted result
  • Editing of the result before export
  • A persistent local library with full-text search
  • Batch conversion, library filters and per-file retry
  • Screenshot capture via Cmd+V
  • A token count displayed on every document

A separate CLI ships the same conversion engine: npm install -g sygal-cli. The vendor says it produces strict JSON output for scripts and is compatible with orchestrators such as Hermes and OpenClaw.

The token-reduction claim

Sygal's main quantitative pitch is token savings. The vendor states that images, scans, PDFs, web pages and YouTube content converted to text can run roughly 2 to 7 times lighter on tokens than the raw source, because pixels, layout and markup are stripped out. The page explicitly qualifies this: the gain is largest on visual and web content, and exact figures vary with each file.

For already-structured formats such as Word, Excel and PowerPoint, the stated benefit is different: standardizing everything into one predictable Markdown and JSON shape rather than reducing token count.

The "local" dispute

Sygal's headline is "100% Local" and "100% on-device," and its privacy section splits data into two columns. Files, converted text, the local library and locally encrypted API keys stay on the machine. Only two things can reach the network: an image or scan sent to a chosen OCR provider, and a URL the user chooses to import.

Commenters argued that this distinction is buried relative to the headline claim. The default OCR path is a cloud provider.

"Seems like it sends your documents to the LLM of your choice. In what world is that 'locally'??" — kennywinker

"How does this differ from open source cli tools such as Marker? Also, it seems disingenuous to have 'locally' in the title when your default is cloud inference providers and I don't see you naming a recommended LLM that could be hosted locally and is proven to work well for this task. You list 'Your files' under 'Stays on your machine' and then require an API key for a cloud provider." — cameronbedard

That second comment raises two distinct points: the absence of a recommended local OCR or vision model, and the comparison to Marker (github.com/datalab-to/marker), an open-source CLI for document conversion.

OCR is bring-your-own-key

Sygal does not bundle an OCR model. Users supply their own API key and pick a provider. The product page lists:

  • OpenAI — the default; billed per token by image resolution; positioned for everyday mixed documents
  • Mistral — described as best value; flat price per page with a cheaper asynchronous batch mode; positioned for large volumes of scans
  • Claude — described as premium and the priciest at volume
  • NVIDIA NIM — includes a free option to start without an API budget
  • Gemini — also available
  • Custom provider — supported

The vendor states there is no markup on OCR and that the provider can be swapped at any time. No locally hosted OCR engine is named on the page.

Pricing and licensing

Sygal is sold as a one-time purchase rather than a subscription:

  • €25 launch price, listed against a regular €29
  • Licence for 2 seats, usable on either the desktop app or the CLI
  • All features included: conversion, OCR (BYOK), library, batch export
  • Updates included for the major version
  • 14-day free trial with no card required
  • Refund available within 14 days
  • No account required, and the vendor states there is no telemetry

A separate team offer exists for multiple licences.

Alternatives raised in the thread

Two alternatives were posted in the comments, both addressing the local-processing concern directly.

"https://github.com/amitxm/AIDocPrep is the one I built that is true local OCR. I made sure there is NO internet use, even accidentally. Also, added redaction." — axm997

"As a service: https://into.md/" — cachius

A third commenter, Benard-dev, said they are also building a web-page-to-Markdown converter and asked how Sygal handles messy scans and tables. No answer to that question appears in the thread.

What is unresolved

The thread contains no visible reply from the developer. The only comment from the submitter account, cyrillsemah, is flagged and shows no content, so the accuracy-versus-marketing objection raised by kennywinker and cameronbedard is not addressed in the discussion.

Two open questions follow from the source material: whether Sygal can be configured to run OCR fully offline, and how its table and messy-scan handling compares to open-source converters such as Marker. Neither is answered by the product page or the comment thread.

Sources