XYolo is an installable Python project for local YOLO training, Docker training, and web-based training workflows. It can be distributed as a wheel.

After installation, the Python environment provides the xyolo command:

xyolo --help
xyolo --version

For Chinese documentation, see README-CN.md.

Install from source:

python3 -m pip install .

Install in editable mode:

python3 -m pip install -e .

Install a built wheel:

python3 -m pip install dist/xyolo-0.1.0-py3-none-any.whl

ultralytics and PyYAML are installed as Python dependencies. XYolo no longer creates or manages a separate venv/ inside the working directory.

Build the frontend assets first:

cd web/ui
corepack pnpm install --frozen-lockfile
corepack pnpm build
cd ../..

Then build the Python distributions:

python3 -m pip install build
python3 -m build

Artifacts are written to dist/. The frontend build writes to src/xyolo/static/; those files are included in the wheel, so Node.js and pnpm are not required when running the installed web service.

XYolo treats the current directory as the project workspace and automatically uses:

models/          # model weights
datasets/        # dataset configs and related files
runs/            # training output
web/configs/     # YAML configs saved by the web UI
web/drafts/      # web drafts
web/templates/   # web templates
web/tasks/       # web tasks and logs

The installed package contains only application code and static web assets, not user training data.

xyolo train model=yolov8s.pt data=dataset.yaml epochs=200 batch=8

--mode venv remains as a compatibility name and is still the default. It now means using the Python environment where XYolo is installed:

xyolo train --mode venv model=best.pt data=my_dataset.yaml epochs=100

Files found under models/ and datasets/ are resolved automatically, and project=runs is added unless explicitly supplied.

Print the resulting command without launching training:

xyolo train --dry-run model=yolov8s.pt data=dataset.yaml epochs=1

xyolo train --mode docker model=yolov8s.pt data=dataset.yaml epochs=200

Docker mode mounts the current workspace at /ultralytics. It uses ultralytics/ultralytics:latest and runs detached by default:

xyolo train --mode docker --attach model=yolov8s.pt data=dataset.yaml
xyolo train --mode docker --container-name train-01 model=yolov8s.pt data=dataset.yaml

The CLI modules follow the web UI sections:

xyolo dataset    # reserved
xyolo train      # implemented
xyolo model      # reserved
xyolo eval       # reserved
xyolo deploy     # reserved
xyolo web        # implemented

Reserved modules are discoverable through xyolo --help. Invoking one currently returns a clear not-implemented message, and future commands can be added under the matching module.

xyolo web
xyolo web --host 0.0.0.0 --port 8860

The server reads the bundled frontend assets from the installed wheel. Task metadata and training output remain in the current workspace.

The web console now uses a 1Panel-style left-right workspace:

• A persistent left sidebar switches between Datasets, Training, Models, Evaluation, and Deploy
• The sidebar can be collapsed and the state is remembered in the browser
• The right content pane stays focused on the active module, and the training module keeps New / List / Template as secondary tabs in the header
• Language, theme, environment details, and external tool entry points stay in the top action area of the content pane

cd web/ui
corepack pnpm install --frozen-lockfile
corepack pnpm dev

The Vite development server proxies /api and /logs to http://127.0.0.1:8860.