Lumina on a Mac
Local-first desktop AI agent with persistent memory & voice · Agent Platforms
Lumina is a local-first agentic harness; a native desktop assistant (Python + PySide6/Qt, no Electron) built for running on your own hardware rather than someone's cloud. It keeps multi-tier persistent memory with temporal decay, so it remembers across sessions, and ships around 70 tools covering the filesystem, sandboxed Python and shell execution, and browser automation, plus subagents for delegated work and scheduled background tasks. It can also write its own tools, clone a voice, and swap personas. It is model-agnostic: llama.cpp is the recommended backend, and it also drives Ollama, LM Studio and vLLM; all of which FrontierStack already manages; or any cloud provider (Claude, Gemini, OpenAI, DeepSeek, Groq, OpenRouter and more) via an OpenAI-compatible endpoint. Install by cloning the repo, pip install -r requirements.txt and playwright install chromium; copy config.example.py to config.py and point it at your backend. Two caveats on a Mac: macOS support is still in testing (Linux is the primary target), and it needs brew install portaudio first for the voice stack. The project targets an NVIDIA GPU with CUDA for local inference, and documents no Apple-silicon/Metal path; so on this Mac, point it at a model served by Ollama, LM Studio or a cloud provider rather than expecting local GPU inference. Apache-2.0.
Run Lumina with FrontierStack
Install or connect Lumina; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks, Malware Audit and backups where they apply.
Run it from your Mac.
FrontierStack installs, monitors and secures services on this Mac and on linked servers.
Download FrontierStack