Pyxis

An all-local AI document reader suite for the ultimate learner.

Pyxis is a local-first PDF reader with multimodal AI built in. It reads pages aloud via neural TTS, describes images and charts for blind users, answers questions about your documents with RAG-grounded citations, and writes WYSIWYG Markdown notes alongside every PDF. The AI model (Gemma 4), the TTS engine (Piper), and your documents all run entirely on your machine — no cloud, no API keys, no telemetry.

Features

• Native PDF rendering via PyMuPDF (MuPDF) with LRU render cache
• WYSIWYG Markdown notes — syntax markers hide visually, inline images render, auto-saves per PDF
• Multimodal AI (Gemma 4) — summarize notes/pages, Q&A with RAG + page citations, extract to-dos, draft follow-ups, suggest tags. Streams output live into notes.
• Accessibility mode 🎧 — Piper neural TTS reads pages aloud; vision model describes images for blind users; descriptions cached for re-opening
• RAG retrieval — BM25 + fuzzy blend over paragraphs/tables/images with budget-aware context assembly
• Highlights & captures — saved as PNGs, embedded into notes
• Cross-platform — Linux (AppImage), Windows (portable EXE)
• Dark theme by default
• No cloud, no API keys, no telemetry

The AI layer needs ≥ 8 GB RAM. On first use it downloads an open-weights Gemma 4 GGUF model (sized to your machine) + a ~65 MB Piper voice file. If you have an NVIDIA GPU, it auto-downloads a CUDA build for faster inference.

Download

Pre-built binaries are on the GitHub Releases page:

• Linux: Pyxis-x86_64.AppImage — single file, no install. chmod +x && ./Pyxis-*.AppImage
• Windows: Pyxis.exe — single portable executable. Double-click to run.

Build from source

Requires Python 3.10+.

git clone https://github.com/Ibzie/Pyxis.git
cd Pyxis
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r requirements.txt

For GPU acceleration (NVIDIA) build llama-cpp-python against CUDA:

CMAKE_ARGS="-DGGML_CUDA=on" pip install --upgrade --force-reinstall llama-cpp-python

On Apple Silicon, Metal is picked up automatically from the prebuilt wheel.

Usage

# Open the welcome screen
python main.py

# Open a PDF directly
python main.py document.pdf

For each opened PDF a folder is created at ~/.local/share/pyxis/notes/<pdf-name>/ (Linux) or %APPDATA%\pyxis\notes\<pdf-name>\ (Windows) containing:

• notes.md — editable Markdown notes
• highlights/ — image snippets of highlighted text
• captures/ — image snippets of screen captures
• annotations.json — metadata index + cached image descriptions

AI menu (toolbar → AI)

Command	What it does
Load AI Model	Detects RAM/accel, downloads a fitting Gemma 4 GGUF, loads in-process
Summarize Notes	Markdown bullet summary of the whole notes.md
Summarize Current Page	Bullet summary of the page in view
Ask…	Free-form Q&A grounded in your PDF via RAG (with page citations)
Extract To-Dos	Markdown checklist of action items
Draft Follow-up	A short connecting note
Suggest Tags	A line of #tag tokens

AI output is streamed live into the notes panel and saved like any other note. Esc cancels an active run.

Accessibility (🎧 toolbar button or Ctrl+Shift+A)

Shortcut	Action
R	Read current page aloud (TTS)
S	Stop narration
Space / P	Pause/resume narration
I	Describe next image on current page (vision model)
N	Read notes panel aloud
Alt+Left/Right	Navigate pages (stops narration first)

Right-clicking an image region triggers description immediately.

General Keyboard Shortcuts

Shortcut	Action
Ctrl+O	Open file
Ctrl+0	Toggle fit-to-width
Ctrl+= / Ctrl++	Zoom in
Ctrl+-	Zoom out
Alt+Right	Next page
Alt+Left	Previous page
Esc	Clear search / cancel capture / cancel AI / stop narration

Tech Stack

Python · PyQt6 · PyMuPDF · llama.cpp · Gemma 4 · Piper TTS

License

MIT — see LICENSE.