homie-ai 0.1.0

Fully local, privacy-first personal AI assistant

Homie AI

Fully local, privacy-first personal AI assistant. Runs entirely on your machine with no cloud dependencies.

Features

• Local inference — loads GGUF models directly via llama.cpp with GPU acceleration (CUDA)
• Privacy first — all data stays on your machine, no telemetry, no external API calls
• Voice pipeline — wake word detection, speech-to-text, text-to-speech (optional)
• Plugin system — 12 built-in plugins (system, clipboard, browser, IDE, git, health, music, notes, shortcuts, terminal, network, workflows)
• Behavioral intelligence — habit detection, routine observation, profile synthesis
• Memory system — working, episodic, and semantic memory with consolidation and forgetting

Quickstart

Install

pip install homie-ai

Setup

# Interactive setup — detects hardware, finds/downloads a model
homie init

# Or start chatting directly (uses default model path)
homie chat

Model Setup

Homie uses llama.cpp server for inference. Download the pre-built binary for your platform from the releases page and place it in ~/.homie/llama-server/.

For Windows with CUDA:

# Download and extract to ~/.homie/llama-server/
# The zip should contain llama-server.exe and required DLLs

Homie auto-discovers GGUF model files in common locations (~/.lmstudio/models/, ~/.homie/models/). You can also register models manually:

homie model add /path/to/model.gguf --name my-model --format gguf --params 35B
homie model switch my-model

Configuration

Homie uses homie.config.yaml in the working directory or ~/.homie/:

llm:
  backend: gguf
  model_path: /path/to/your/model.gguf
  context_length: 65536
  gpu_layers: -1       # -1 = offload all layers to GPU
  max_tokens: 2048
  temperature: 0.7

voice:
  enabled: false
  wake_word: "hey homie"
  mode: push_to_talk

storage:
  path: ~/.homie

Environment variable overrides: HOMIE_LLM_BACKEND, HOMIE_LLM_MODEL_PATH, HOMIE_LLM_GPU_LAYERS, HOMIE_VOICE_ENABLED, HOMIE_STORAGE_PATH, HOMIE_USER_NAME.

CLI Commands

homie start          Start the assistant (alias for chat)
homie chat           Interactive chat mode
homie init           First-time setup wizard
homie model list     List installed models
homie model add      Register a local model file
homie model switch   Switch the active model
homie plugin list    List available plugins
homie plugin enable  Enable a plugin
homie backup --to    Create encrypted backup
homie restore --from Restore from backup

Optional Dependencies

pip install homie-ai[model]    # HuggingFace model downloading
pip install homie-ai[voice]    # Voice pipeline (STT, TTS, wake word)
pip install homie-ai[context]  # System context tracking
pip install homie-ai[storage]  # Vector DB and encrypted backups
pip install homie-ai[app]      # Dashboard, system tray, scheduling
pip install homie-ai[all]      # Everything

Requirements

• Python 3.11+
• llama.cpp server binary (b8149+ for Qwen3.5 support)
• A GGUF model file (auto-detected or manually configured)
• NVIDIA GPU recommended (CUDA); CPU-only mode is supported

License

Mozilla Public License 2.0