Gloss

Annotate as you read — highlight a passage, attach a note
Reader — inline and display LaTeX, with an outline rail
Export — a structured packet, framed for your LLM
Library — your documents, stored locally

Gloss is a desktop reader for annotating documents — lecture materials, research papers, LLM outputs, etc. — and exporting your highlights and notes as a single structured packet you can hand to an LLM.

When you want ChatGPT / Claude to revise or explain several specific parts of the output, the UI only allows you to directly reply to one selected piece of text at a time. More generally, if you’re using Claude Code / Codex or similar, you have to type out all your questions / requests in the chat box, rather than being able to annotate the model’s output directly.

Gloss is built to fix this: open a markdown, LaTeX, or PDF file, highlight the passages you care about, attach a note (or dictate one by voice), and export everything—your highlights, their surrounding context, and your questions—as one block of markdown text. Paste it into Claude Code, Codex, or your favourite LLM, and the model will respond to each item in order. If you use Claude Code / Codex, you can easily get it to update the relevant document in-place to address your annotations.

Download for Mac Download APK GitHub

Everything after this point is AI-generated.

Features

  • Annotate as you read — select any passage and type or dictate a note. Highlights stick to the text, even across inline math.
  • Built for real papers — renders inline and display LaTeX; imports markdown, .tex (via pandoc), and PDFs (via Mathpix).
  • Export a structured packet — your annotations plus surrounding context, framed for the model. Use the default instruction or write your own, and choose how much context to include.
  • Dictation-friendly — your keystrokes go straight into the note, so any dictation tool (Wispr Flow, macOS Dictation) lets you speak an annotation instead of typing it.
  • Local-first and private — no accounts, no telemetry, no network calls except a PDF conversion you trigger yourself.
  • Android companion — read and annotate markdown on the go.

How it works

  1. Import a markdown, LaTeX, or PDF file — or paste text.
  2. Annotate the parts you want to ask about.
  3. Export the packet and paste it into your LLM.
Install
  • macOS — download the .dmg from Releases, drag Gloss.app to Applications, and open it. The build is unsigned, so on first launch use right-click → Open. Apple-silicon Macs only.
  • Android — a markdown-only companion, distributed as a debug APK for now. Install with adb install -r Gloss-android-debug.apk.
  • Or build from source (JDK 17 + Gradle).
Optional helper tools (desktop)

Markdown and plain text work with no setup. To import other formats:

  • .tex — install pandoc (brew install pandoc), runs locally.
  • .pdf — install the Mathpix mpx CLI, or paste a Mathpix API key in Preferences. PDF OCR uploads the file to Mathpix; nothing else leaves your machine.
Privacy
  • Everything stays local by default — documents and annotations live in a local database.
  • No analytics, no telemetry, no background network calls.
  • The only network call is PDF conversion via Mathpix, and only when you actively import a PDF on desktop.
  • See the full privacy note.

Open source under the MIT license. Feedback and issues welcome on GitHub.