Localis

A local AI assistant that runs entirely on your machine. No cloud, no subscriptions — your data stays private.

Features

• Local LLM inference via llama-cpp-python (GGUF models, GPU-accelerated)
• Agentic tool-calling — single-model loop with 7 built-in tools
• Persistent memory — remembers facts about you across conversations
• RAG — upload PDFs, Word docs, and CSVs for in-context retrieval
• Web search — Brave or Tavily integration (optional)
• Notes & Reminders — voice or chat-triggered, with timed pings
• Finance Advisor — upload bank/credit CSVs for spending analysis
• Home Assistant — control smart home lights and devices
• Voice — wake word ("Hey Jarvis"), Whisper STT, Piper TTS (optional)

Requirements

• Python 3.12
• A GGUF model file (e.g. Qwen3.5 from HuggingFace)
• NVIDIA GPU recommended (CUDA); CPU-only works but is slow

Installation

git clone https://github.com/Rishi943/Localis.git
cd Localis

python3.12 -m venv .venv
source .venv/bin/activate

pip install -r requirements.txt

uvicorn app.main:app --host 0.0.0.0 --port 8000

Open http://localhost:8000. A setup wizard will guide you through downloading a model on first launch.

Configuration

Create a secret.env file in the project root:

MODEL_PATH=/path/to/models/directory

# Web search (optional)
BRAVE_API_KEY=your_key
TAVILY_API_KEY=your_key

# Home Assistant (optional)
LOCALIS_HA_URL=http://homeassistant.local:8123
LOCALIS_HA_TOKEN=your_long_lived_token
LOCALIS_LIGHT_ENTITY=light.your_light

Voice Support (Optional)

Voice requires Python 3.11 due to tflite-runtime compatibility.

bash scripts/setup_voice_venv.sh

You also need the Piper TTS binary in your PATH and a voice model configured via LOCALIS_PIPER_MODEL in secret.env.

Data

All persistent data is stored at ~/.local/share/localis/ — database, models, and uploaded files.