What it is
LocalXiv is a Mac app for reading research papers. When you paste an arXiv or alphaXiv link, the app saves the paper to a library on your machine. The reader renders the equations and lets you set the font, text size, and margins. You can export a paper as an EPUB and send it to a Kindle.
The AI features are optional. An Overview explains the paper's contribution visually, and a Blog gives a longer explanation that cites the source. You can also ask questions in a chat. These features use your own AI provider and API key. You can import, read, and export papers without an AI account.
Why it was needed
An AI explanation of a paper is useful only if you can check it. LocalXiv keeps the original paper beside every explanation, so you can check each claim against the source.
What I built
- The app has three parts: a Swift and AppKit launcher, a WebKit reader, and a local Python service. SQLite stores the library and the job records.
- Paper conversion and AI generation run as separate jobs, so a failed explanation does not affect a saved paper. If EPUB conversion fails, the app uses the saved original instead.
- The model does not draw the Overview figures. It writes a digest and a scene tree of cards, groups, steps, and bars. The app then places those parts and routes the arrows. If a figure still fails after a limited number of correction attempts, the app leaves it out.
- I developed LocalXiv mostly with GPT-6 Astra in Codex. Each user picks the model that runs inside the app.
What it achieved
- I shipped twelve public preview releases, v0.0.1 to v0.0.14, between 6 and 18 September 2026.
- LocalXiv launched on Product Hunt as an entry in the GPT-6 Astra Challenge.
- The app ships its own conversion tools, so you can import and read papers without Homebrew or Python.
Status
LocalXiv is in early preview and runs on Apple Silicon Macs with macOS 26 or newer. I sign releases ad hoc, and Apple has not notarized them yet. Conversion can be imperfect for unusual equations, tables, and layouts. AI explanations can contain mistakes.
Related
- Verification notes: SVG figure reliability (opens in a new tab)
- Xiv2Pub: arXiv papers to EPUB (opens in a new tab)
- VisLearn-Paper: a visual paper interpreter (opens in a new tab)
Send questions or feedback to akhiltvsn@gmail.com.
