feedIO

A desktop RSS/Atom reader with offline reading, topic-specific voting and learned ranking. This is the first beta of version 2.0.

Run from source

Requires Python 3.10–3.14 and a desktop capable of running Qt 6 WebEngine. See platform status for tested configurations and known limitations. Create and activate a virtual environment:

python -m venv .venv
# Linux/macOS:
source .venv/bin/activate
# Windows PowerShell instead:
# .venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install .
python -m feedio

feedio is also installed as a desktop entry command. Use python -m feedio --profile /path/to/new-profile --check to initialize an isolated profile without opening a window; omit --check to use it interactively. Each running instance needs its own profile. The default profile uses the OS user-data directory (platformdirs); FEEDIO_PROFILE overrides it.

Everyday use

Add an RSS/Atom or website URL from Feeds, or import OPML, then refresh. Selecting an article marks it read. Votes train only the selected interest. Bookmarks are local and included in profile backups. Article HTML is sanitized; scripts and remote resources are blocked. Original links open in your system browser. Preferences control periodic refresh, notifications, close-to-tray and speech. Closing exits by default; tray behavior depends on desktop support. Use Cancel in the status bar to stop a refresh or ranking operation. Article → Next/Previous article page reaches older articles throughout the library.

Speech is optional: python -m pip install '.[speech]'. It also requires an OS speech backend (for example eSpeak on Linux). Reading works without a speech backend. Speech is not included in the default desktop bundle. Pocket and authenticated Twitter integrations have been removed. Translation is currently unavailable.

Existing users: read migration and recovery before importing your profile. Old CRM/pickle training files are never executed or loaded.

Learning and ranking

feedIO learns locally from up and down votes. Each interest has its own small text preference model, built from article titles and cleaned article bodies. Scores combine that text evidence with the article's feed history. The model is stored as vote records in SQLite and rebuilt when needed; it does not use CRM114, downloaded training corpora, pickle files or a separate machine-learning service.

Development and distribution

python -m pip install -e '.[dev]'
python -m pytest -q
python -m ruff check .
python scripts/generate_ui.py
python -m build
python scripts/build_desktop.py

See the contribution guide, test commands, architecture, and build instructions. GPL-3.0-or-later; see license and authors.

For graphics-driver problems, try python -m feedio --software-rendering. This turns off GPU acceleration while keeping the browser sandbox enabled.