Lena
A self-hosted AI platform that started as a local model runtime and grew into a smart-home and security assistant.
Lena began as Lena-Runtime: a TypeScript server that runs local language models, with a built-in web interface for chat, sub-agents, MCP tool servers, cloned voices and browsing models on Hugging Face. It deploys with Docker on a plain CPU or an NVIDIA GPU.
From there it grew into a whole system on one Mac mini. Lena Core routes requests between local and cloud models and manages everything else. A dashboard runs the house, and a camera pipeline watches the security cameras, runs a two-stage vision check, recognizes faces and plates, and sends an iMessage alert with a short clip.
A companion tunnel exposes my Claude and OpenAI subscriptions as standard API endpoints, so any tool on the network can use them. There are also chat clients for iPhone, Apple Watch, Apple TV and Mac.
Lena is on pause while Camera Bridge takes most of my time.
Build log
Cortex v2 memory
Structured long-term memory landed, right after a security audit that closed injection, SSRF and path-traversal holes.
One API for every model
Added the API tunnel: Claude and OpenAI subscriptions exposed as OpenAI- and Anthropic-compatible endpoints, with an optional public tunnel.
Lena-Runtime takes shape
First commit of the TypeScript runtime. Thirty-two commits over the next two days added the web UI, sub-agents, MCP servers and voice.