mayanklahiri/virtualme repository overview

⁠Virtual Me

A private background agent with a local LLM, a real browser on a virtual desktop, recurring projects, durable jobs and notifications, speech, outbound mail, and an optional Telegram Bot API integration. Core inference and state stay on your machine; enabling Telegram deliberately sends authorized chat traffic through Telegram's cloud service.

⁠Run it

npx virtualme start

or with plain Docker:

docker run -d --name virtualme -p 8080:8080 \
  -v ~/.virtualme:/home/virtualme/.virtualme \
  mayanklahiri/virtualme:latest

Then open http://localhost:8080⁠ in a browser on the same network.

Your data (browser profile, chat history, projects, configuration, notifications, metrics, and mail queue) lives in ~/.virtualme on the host and survives updates. The console includes Config, Jobs, Data, Notifications, and Telegram integration pages.

⁠Good to know

• Needs about 8 GB of RAM; the local model alone uses ~4 GB.
• With an NVIDIA GPU and the NVIDIA Container Toolkit installed, npx virtualme start passes the GPU through automatically and runs the model on it (--no-gpu opts out).
• First start is slow while the model loads. Give it a few minutes.
• Prototype trust model: no auth or TLS. Run it on a trusted private network only.

Source, docs, and issues: https://github.com/mayanklahiri/virtualme⁠