TeX64 is a desktop LaTeX editor for macOS. You open a folder on your own machine, edit .tex files, build with one click, and read the result in an integrated PDF viewer — a fully local editing loop, augmented with visual math input, image-to-LaTeX OCR, and a LaTeX-specialized AI (Axiom). This page is a plain tool-guide entry: what each feature actually does, and how.
What TeX64 is
TeX64 is local-first: apart from its AI features it runs entirely offline, and both your source and your PDF stay on your own machine. It currently ships for macOS only (macOS 12 Monterey or later), with native support for both Apple Silicon (M1/M2/M3/M4) and Intel Macs. Windows and Linux builds are planned but not yet available — distribution is paused while signing and installer policy are finalized.
The shape of using it is simple. Open a workspace (a folder) and you get an editor with a file tree, tab editing, and split view. Your main .tex file is auto-detected, and pressing Cmd+B builds it, showing the PDF in the integrated viewer. The editing essentials are all present: an outline auto-generated from chapters, sections, figures, labels, citations, and TODOs, and full-text search across the whole workspace (powered by ripgrep).
Note that TeX64 is not itself a typesetting engine. Compilation runs a TeX distribution you install separately (see below) **through latexmk**. In other words, TeX64 is the front end that ties editing, building, and preview together, while the typesetting itself is done by the standard TeX toolchain.
Installation
Installing the app itself is the usual macOS routine. From the download page (/download), grab the DMG that matches your machine (Apple Silicon or Intel), open it, and drag TeX64.app into your Applications folder. It needs roughly 150 MB of disk space.
On first launch, macOS Gatekeeper may show a warning; if so, Control+click the app and choose “Open” to start it. If you are unsure whether your Mac is Apple Silicon or Intel, run uname -p in Terminal (arm means Apple Silicon, i386 means Intel). The DMGs are signed and notarized, and a checksum is published per artifact.
TeX environment setup
As noted, building in TeX64 requires a TeX distribution installed separately. The recommendation is MacTeX (full TeX Live plus GUI tools, ~4 GB). To keep things lean, you can instead install BasicTeX (~100 MB) and add missing packages over time with sudo tlmgr install.
The tools TeX64 specifically needs are lualatex (the default engine), latexmk (which drives the build), latexindent (for formatting), and synctex (for SyncTeX). You can check whether they were found under Settings > Environment in the app: a green check (✓) means found, red means missing. If something is missing, install MacTeX or add the individual package with tlmgr.
For the case where no TeX environment is present at all, TeX64 can also install a managed TeX Live for you (added in version 0.1.12): when the required tools are missing, the app can set up a TeX Live environment under its own management.
- Recommended: MacTeX (full TeX Live + GUI, ~4 GB)
- Lean: BasicTeX (~100 MB) plus
tlmgrfor the rest - Verify: green checks for
lualatex/latexmk/latexindent/synctexunder Settings > Environment
Math input (Blocks)
Blocks (the visual math editor) is the feature for building equations without memorizing LaTeX notation. Under the hood it uses MathLive, an interactive equation builder: you pick from a palette of Greek letters, operators, relations, arrows, set symbols, structures such as fractions and roots, and accents to assemble the formula. As you type you see the rendered result in real time, and when you are done the corresponding LaTeX code is inserted back into your source automatically.
There are three ways in: place the cursor in a math environment and right-click > “Edit in Block,” use the menu Math > Block Editor, or press **Cmd+Shift+E**. TeX64 detects math environments automatically and launches the block for them. Covered environments include the displayed ones — equation, align, gather, multline, split, cases — the matrix family, and inline math written as $...$ or \(...\).
Math OCR (image → formula)
OCR turns the math in an image into LaTeX code. It works on screenshots, handwritten notes, formulas inside PDFs, and photos of equations from textbooks or papers — handwriting and print alike. To use it, choose Math > OCR from the menu, crop any region of the screen, and the AI analyzes the image, generates the LaTeX, and inserts the result straight into the editor.
Because conversion runs on cloud AI, it needs an internet connection and an AI-enabled plan (the docs describe it as an AI feature on TeX64 Basic or higher). And because it captures a region of the screen, you may need to enable TeX64 under macOS Screen Recording permission (System Settings > Privacy & Security > Screen Recording). For better accuracy: use high-contrast images (dark text on a light background), crop to just the equation, write handwriting as neatly as possible, and split complex formulas. Always review and correct the output before relying on it.
As a rule of thumb: reach for Blocks when creating a new formula from scratch or when LaTeX notation is unfamiliar, and OCR when digitizing an existing handwritten or PDF equation.
AI writing (Axiom)
Axiom is the LaTeX-specialized AI agent built into TeX64. It understands the context of your whole project and does more than chat — it reads and writes files directly as it works. There are two entry points: AI chat, where you talk in natural language (“Why is this error happening?”, “Make this section more concise”), and inline completion, which suggests the next line, equation, or paragraph in the editor. Completion can also turn natural language such as “integral of x squared from 0 to 1” into an equation like \int_0^1 x^2 dx.
As an agent, Axiom reads your project files (.tex, .bib, .sty, .cls, and so on) and the file tree, and proposes edits as diffs. Crucially, those diffs are always applied only after you review and approve them. It can also compile the project and read the build log, and search arXiv to fetch BibTeX entries. On the other hand, it cannot reach files outside your workspace and cannot run arbitrary system commands (its tools are restricted); API keys and tokens are redacted automatically.
There are two models: the standard Axiom model is available on every plan, and the most capable Axiom Pro is Pro-only. On privacy: your paper data is never used for AI training, data is sent only when you explicitly use an AI feature, it is processed server-side via the OpenAI API, and it is deleted after processing rather than stored long-term. The plans are Free, Basic, and Pro, with higher tiers offering more capable models and larger monthly AI allowances (see the pricing page for exact figures). Always verify the output — Axiom is powerful but can be wrong about complex code or semantics.
SyncTeX (forward/inverse search)
SyncTeX links matching positions in source and PDF in both directions, and it is built into the TeX64 PDF viewer. With forward search (source → PDF), the area of the PDF corresponding to your cursor is highlighted and the viewer scrolls there automatically. With inverse search (PDF → source), **Cmd-clicking** the PDF jumps straight to the corresponding source line and highlights it (for multi-line blocks such as a paragraph or a figure, it jumps to the start of the block).
For this to work, synctex must be available and the build must generate a .synctex.gz file. TeX64’s build enables .synctex.gz generation by default, so it usually works with no extra setup. It also works in multi-file projects: click an element from a file split out with \input or \include, and the correct external file opens automatically and jumps to the right line. That assumes the root file (the main file containing \begin{document}) is set correctly — TeX64 tries to auto-detect it, and you can set it explicitly via the menu’s “Set Root File” when needed.
Jump precision depends on how granular the information TeX64 can read from the SyncTeX file is. Often you land on the right line at paragraph-level accuracy, but some packages such as tikz or pgfplots, and heavy macro use, can complicate the mapping and yield coarser (for instance, page-level) precision.