The 6 PDF readers I tried before building my own

I spent about a year trying every PDF reader pitched at researchers before deciding to build Voixes. This is a partial honest accounting of what I learned.

Three things up front:

1. None of the tools below are bad. Most of them are good at the thing they were built for.
2. I'm clearly biased — I built a competitor. Read accordingly.
3. The tool you actually use depends on what you're trying to do. There is no "best PDF reader" in general, only "best PDF reader for this kind of work."

What I was looking for

My setup: I read 5–15 papers a week across machine learning, applied math, and adjacent fields. I want to be able to:

• Read the paper without distraction. Sidebars, popovers, chat interfaces that hijack my attention all fail this.
• Highlight, annotate, and have those marks survive across sessions and devices (ideally without forcing me into a sync account).
• Quickly find a figure or table without scrolling.
• Follow citations without losing my place.
• Ask basic questions about the paper when I'm too tired to read the full intro carefully.
• Pull data out of a plot when I need it.

Nothing I tried hit all of these.

Preview.app

The Mac default. It is genuinely good at the basics: opens fast, scrolls smoothly, highlights work. For 80% of casual reading it's all you need.

What it doesn't do: any AI, any figure extraction, any citation handling beyond "click and your browser opens." For deep reading of unfamiliar fields it stops being enough by paper 3 of any new topic.

Mendeley / Zotero / EndNote class

These are reference managers first, readers second. The reading experience inside them is usually adequate but never the focus. Zotero is the strongest of the three IMO — the highlighter is fast, the notes panel is useful, and the library organization is genuinely excellent.

What's missing: figure extraction is nonexistent, "ask the paper" is bolted on if it exists, and plot digitization isn't even on the roadmap. They're great if you're managing a 2000-paper library; less great if your bottleneck is reading the next 5 papers carefully.

ReadCube Papers

Beautiful interface. Has some AI features now. Cloud-first by default — your library lives on their servers — which is a dealbreaker for some.

The strongest thing about ReadCube is the visual organization: tags, related-papers suggestions, the discovery feed. The weakest is the price for what you get, especially as a student.

ChatPDF / HelloPDF / generic "ask the paper" tools

This category exploded in 2023–2024. They all work approximately the same way: upload a PDF, get a chat interface, ask questions, get cited answers.

What they get right: the chat answers are surprisingly good. Citation linking back to the source paragraph is usually accurate. They're zero-friction for "quick question about this paper I don't really want to read."

What they get wrong: everything else. They're not built for deep reading — they're built for skimming. The PDF view is usually a thin shell over PDF.js. No figure extraction, no annotation export, no offline mode, and the PDF lives on their servers.

If you only ever skim, they're fine. If you actually need to read anything carefully, they're not the tool.

Highlights (for Mac)

A native Mac reader focused on highlighting and exporting notes. Lovely typography, excellent export to Markdown/Bear/Obsidian. Worth knowing about if you're a heavy note-taker.

What's missing for me: no AI, no figure extraction. You're getting a really good highlighter, not a reading environment.

What I kept wanting

After a year of cycling through these, the things I wanted that nothing did well:

1. A figure board. Pull every figure and table in the paper into one grid view so I can find Figure 3 without scrolling 18 pages. This is the genesis of Voixes' Quick Glance.

2. Ask, but grounded. I want to ask "what does Figure 4 show?" and get an answer that cites Figure 4. Not "this paper proposes a method for X." If the answer doesn't cite where it came from, it's worthless.

3. Inline citation cards. Click any [12] and see the citation right where I clicked, not in a separate window. AI summary of what the cited work is about. Critically, this needs to work offline because most of the time I'm reading on a plane or in transit.

4. Plot digitization built in. I shouldn't have to copy a figure into WebPlotDigitizer to recover numbers. It's the same workflow as reading — keep it in the same app.

5. No upload, ever. I read confidential drafts for friends. I read papers I'm reviewing. None of these belong on someone else's server.

That list became the spec for Voixes. Whether I succeeded is a separate question, but at least I knew what I was building.

The honest recommendation

If you're a student or early-career: Zotero + the chat tool of your choice is the cheapest stack that works.

If you read a lot of papers in a single field and you're in front of your Mac: try Voixes. (Obviously.)

If you only need to skim and you don't care about privacy: ChatPDF or its competitors are fine.

If you're in a sensitive field — clinical, legal, security, anything pre-publication — anything cloud is the wrong answer. Use Highlights or Voixes.

---

Voixes is a native Mac app for reading research papers — figure board, citation cards, plot digitization, ask-the-paper. All local. Free to start.