Cross-platform bootstrap for a managed
.venvand frontend dependencies.
Source: scripts/setup.py
User guide: Installation
The setup script performs prerequisite checks, creates .venv with Python 3.10,
installs the selected editable extras with uv, installs locked frontend
dependencies, validates the result with Doctor, and builds the frontend.
--check stops after the non-mutating prerequisite pass.
| Profile | Python extras | Additional work |
|---|---|---|
local-gpu |
all,dev |
Install LatentSync sidecar and download Qwen/WhisperX models |
remote-client |
client,web,dev |
Optionally validate a remote GPU server |
dev |
client,web,dev |
Development environment without local GPU models |
auto |
Resolves to one of the profiles above | Select local-gpu only on Linux/WSL2 with an accessible NVIDIA GPU; otherwise select remote-client |
Native Windows and macOS reject local-gpu; Windows users can use WSL2 for the
local CUDA stack. Profile resolution intentionally matches Doctor.
| Option | Default | Description |
|---|---|---|
--profile auto|local-gpu|remote-client|dev |
auto |
Installation capability profile |
--model 0.6B|1.7B |
0.6B |
Qwen model downloaded and checked for local-gpu |
--server-url URL |
none | Remote GPU server passed to the final doctor run |
--check |
false | Check prerequisites without creating or changing the environment |
python3 scripts/setup.py --check
python3 scripts/setup.py --profile auto
py scripts/setup.py --profile remote-client --server-url http://gpu-vm:8100
The prerequisite pass requires Git, uv, Node/npm, ffmpeg, and ffprobe. The local
GPU profile additionally requires the compiler toolchain, Bash, and a working
nvidia-smi command. Failures include platform-specific installation guidance
and return a nonzero exit code.
.venv with uv venv --python 3.10.npm ci when package-lock.json exists.local-gpu, install LatentSync and preload Qwen/WhisperX models.