Local-first AI system

MAUDE

A private AI command layer running on the DGX Spark, built to connect models with real tools while keeping work inspectable.

Install MAUDERead the case study

Install MAUDE

git clone https://github.com/mboard8070/terminal-llm.git
cd terminal-llm
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
playwright install chromium
./setup_local.sh
./maude
189

server tools

5

client surfaces

local-first

gateway design

Install

Run MAUDE on a Linux workstation with CUDA.

The server install is intended for Ubuntu 22.04+ or a similar Linux machine with an NVIDIA GPU, CUDA drivers, Python 3.12+, Git, cmake, make, g++, tmux, and curl.

Full setup guide

Server install

git clone https://github.com/mboard8070/terminal-llm.git
cd terminal-llm
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
playwright install chromium
./setup_local.sh
./maude

Client CLI

pip install --upgrade "git+https://github.com/mboard8070/terminal-llm.git#subdirectory=maude-client"
maude-client

Use the client after your MAUDE gateway is running and reachable, typically over Tailscale.

What it does

A practical assistant stack, not a chatbot demo.

Routes local and cloud models through one private gateway
Connects chat to files, shell, browser automation, GitHub, Google Workspace, media tools, and scheduled missions
Keeps sensitive actions visible with traces, approval points, and human-readable logs