Personal Reasoning Autonomous Distributed Agent. A JARVIS-inspired assistant for the home that runs entirely on your own hardware: no cloud, no accounts, no API keys. A home node is the brain; you talk to her by voice or in the terminal, and she picks the right module to answer. She never claims to have done something she didn't: actions are checked in code before she says they're done.
Early construction, built for one home first.
- Talk by voice: wake word, offline speech recognition and a local voice. Say the wake word again to interrupt her.
- Teach languages: a Japanese, French and Russian tutor that corrects your grammar and keeps the conversation going.
- Remember: notes you ask her to keep, and the conversation you're having.
- Search the web when you ask, through your own SearXNG, and tell you where the answer came from.
- Play your music: "play Muse Space Dementia" from your own files in Corpus, or on Spotify when it isn't there, in about a second; genres and moods ("something calm") and Corpus' queue too, and pause, resume, skip or ask what's playing in Spotify or any player that shows up in Windows' media controls.
- Wake the PC over the network, and start herself when it boots.
You need Python 3.11+ and Ollama.
ollama pull gemma4:12b
ollama pull nomic-embed-text # her memory search by meaning
python -m venv .venv && source .venv/bin/activate
pip install -e .
prada node # the home node, on 127.0.0.1:7723
prada chat # talk to her in the terminal (works without a node too)For voice, install the voice extra where your microphone is, download the models once, and listen:
pip install -e ".[voice]"
prada voice setup
prada voiceSettings live in ~/.config/prada/config.toml (or $PRADA_CONFIG), and who she is (name, character, tone) lives in persona.toml next to it. Both stay on your machine, never in the repo. prada persona init writes a persona to edit; without one she uses a short neutral default.
docs/setup.md covers the rest: starting at boot on Windows, sleep and Wake-on-LAN, voice settings, training your own wake word, music (and the Spotify login), web search, choosing the model, evals and development.
Next up: a conversation partner rather than a corrector, directed drills, learning your taste in music, and more modules (anime, home devices).