Deployment

mold supports multiple deployment modes; from a single GPU machine to cloud GPU providers.

Remote Rendering

The simplest setup: run the server on a GPU host, generate from anywhere.

# On your GPU server
mold serve
mold host invite                 # prints a one-use pairing link

# Paste the link on your laptop, then generate there
mold host add                    # then paste the link
mold run "a cat" --host workstation

Secure is the default for both direct connections and the optional proxy. Hosts can also opt into Regular direct HTTP(S) for REST apps. See Secure and Regular connections, including persistent app-key generation.

Deployment Options

Method	Best For
Docker & RunPod	Cloud GPUs, RunPod pods
mold runpod CLI	Integrated pod and volume control
mold lambda CLI	Private Lambda Cloud web UI
Secure and Regular connections	Pair clients or issue a direct REST app key
NixOS	NixOS systems, declarative setup
Systemd service	Any Linux with NVIDIA GPU

Systemd Service

Two sample units ship in contrib/. Pick one.

System-wide (contrib/mold-server.service), for a dedicated service user:

sudo cp contrib/mold-server.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable --now mold-server

User mode (contrib/mold-server.user.service), for a single-user GPU box:

cp contrib/mold-server.user.service ~/.config/systemd/user/mold-server.service
sudo loginctl enable-linger "$USER"   # survive logout; once per box
systemctl --user daemon-reload
systemctl --user enable --now mold-server

Key settings:

[Service]
ExecStart=/usr/local/bin/mold serve --port 7680 --bind 0.0.0.0
Environment=MOLD_LOG=info
# NixOS with a non-flake `/usr/local/bin/mold`: uncomment for CUDA driver access
# Environment=LD_LIBRARY_PATH=/run/opengl-driver/lib