Every new paper in your Zotero folder becomes a bilingual PDF and an atomic Obsidian note. Layout preserved. LaTeX intact. No copy-paste. No manual formatting. No middleman.
Free forever with your own API key · Your data stays local
Three steps. One-time setup. Then forget about it.
Point Sulix Cite at your Zotero attachment folder and your Obsidian vault. Takes one minute.
Add a PDF to Zotero like you always do. Sulix Cite detects it instantly.
A bilingual PDF lands in Zotero. An atomic note lands in Obsidian. Both linked. Always synced.
Everything you need. Nothing you don't.
Point Sulix Cite at your Zotero folder. Every new PDF processes automatically. No manual steps. No daily clicks.
Original on the right, translation on the left. Paragraph by paragraph, perfectly aligned. Two-column layouts handled correctly. Citations preserved.
∂ρ/∂t + ∇·(ρu) = 0 stays as ∂ρ/∂t + ∇·(ρu) = 0. Inline math, display equations, nested fractions — all reconstructed pixel-perfect from source. No garbled symbols. Your math stays math.
Every paper becomes a structured Obsidian note. Abstract, key claims, references — all linked. Click sulix:// links to jump back to the source. Your knowledge graph grows automatically.
Sulix Cite never sees your papers. You bring your own API key — translation goes directly from your machine to Anthropic / OpenAI / DeepSeek. No middleman. No storage. No training data.
Process entire conference proceedings in one batch. Resume from interruption automatically. 500-page survey papers? No sweat. Industrial-grade reliability for industrial-scale reading.
Three scenarios. One workflow. Zero friction.
Line-by-line alignment · Confidence audit marks spans needing review
where attention scores a_{ij} are
computed as:
a_{ij} = exp(e_{ij}) / ?
? = ?_k exp(e_{ik})
This ensures the weights sum to 1.
where attention scores \(a_{ij}\) are
computed as:
$$a_{ij} = \frac{\exp(e_{ij})}{\sum_{k} \exp(e_{ik})}$$
This ensures the weights sum to 1.
Advanced ML-powered parsing engines recover every equation to valid LaTeX
---
title: "Attention Is All You Need"
source: [[zotero/vaswani2017]]
---
## Abstract
The dominant sequence transduction models
are based on complex recurrent or
convolutional neural networks ...
$$ \text{Attention}(Q,K,V) = ... $$
One-click export to your Obsidian vault. Frontmatter, citations, and formulas preserved.
Free is genuinely useful. Pro is genuinely worth it. Half off for .edu emails — because we remember being grad students.
For getting started.
$0
Unlimited, forever.
For active researchers.
$49
per year · $24.50 with .edu email
For the long haul.
$129
once · $64.50 with .edu email
Built on BYOK philosophy. We sell the engine, not the fuel. Your API key, your data, your machine. No usage caps. No quota games. No telemetry.
Be the first to know when Sulix Cite ships. Plus: 25% lifetime discount for early supporters.
No spam. One launch email, plus occasional dev updates. Unsubscribe with one click.
2-minute survey. Your input directly shapes our roadmap.
Your responses are anonymous. 3 quick questions.
Zotero is the recommended workflow, but not required. You can drop a PDF directly into Sulix Cite and get the same bilingual output. Zotero just makes it automatic.
No. Sulix Cite works fine without Obsidian — you will still get bilingual PDFs in your Zotero attachments. But if you do use Obsidian, that is where the magic multiplies. Your reading becomes a knowledge graph.
Never to us. Sulix Cite runs entirely on your machine. Your API key talks directly to Anthropic / OpenAI / DeepSeek — we are not in the middle. Your PDFs, your notes, your translations — all local. We literally cannot see them, even if we wanted to.
At launch: English ↔ Chinese (Simplified), with Japanese, Korean, German, Spanish, French in roadmap. Translation quality depends on your chosen LLM — Claude / GPT-4 / DeepSeek all work well for academic text. We are optimizing prompts specifically for academic writing.
For both Free and Pro tiers — yes, BYOK is the model. Your key, your bill, your control. This is by design: it keeps margins clean and gives you complete data sovereignty. A typical paper costs $0.01–$0.03 using Claude Haiku.
Yes. Sulix Cite handles standard academic PDFs (IEEE, ACM, Springer, arXiv, Nature, SAE, etc.). For exotic formats (heavily scanned papers, OCR-only PDFs, non-Latin scripts in body text), results vary — the Pro engine handles these much better than the free fallback.
DeepL translates words. Sulix Cite preserves documents. We keep your two-column layouts intact, your LaTeX equations untouched, your figure captions in place, your references linked. Then we deliver both a bilingual PDF and a structured Obsidian note. DeepL gives you a wall of text. We give you something you can read.
You can — and that is literally what Sulix Cite does under the hood (via your API key). What we add is the pipeline around it: PDF parsing, formula extraction, layout reconstruction, bilingual PDF generation, and Obsidian integration. It is the difference between raw flour and finished bread.
Everything you exported stays yours, forever. Your sulix:// links keep working. Your Obsidian notes keep working. Your bilingual PDFs keep working. You only lose access to new Pro features — but you never lose what you have already built. We do not hold your data hostage.
One person — a CFD engineer turned indie developer, working in the open under Sulix Labs. Built for researchers who are tired of copy-pasting their lives into translation tools. No VC pressure, no growth hacking, no dark patterns. Just a tool that does one thing well.