CLI Graph Runner

You can execute any graph directly from the command line without starting the server. This is handy for batch runs, CI, or reproducing a pipeline headlessly.

If you want to call a saved graph on a running server instead — declared inputs in, declared outputs out, over HTTP — see Graph as a Function. To queue a graph file on a running server, so the run outlives the terminal and appears in the Runs panel, use cdui run; see Run Queue.

run_graph.py imports the backend, so it runs in the backend's virtual environment (backend/.venv, which cdui install creates):

cd backend
source .venv/bin/activate    # Windows: .venv\Scripts\activate
python run_graph.py ../examples/Usage_Example/CNN-MNIST/TrainCNN-MNIST/graph.json

The runner discovers all nodes via the registry, validates the DAG, executes it topologically, and prints per-node output summaries.

Options

Flag	Effect
--validate-only	Validate the graph (DAG, types, ports, Start node) without executing it.
--verbose, -v	DEBUG-level logging, plus the full traceback when a node fails at runtime. There is no CLI switch for the Inspector's step traces.
--device	cpu / cuda / cuda:N / mps / auto. Omitted: the graph file's settings.device, else cpu. auto takes the best accelerator present.
--seed N	Seed every node from N so the run is reproducible. Omitted: the graph file's settings.seed, if any. A seeded run executes one node at a time — see Reproducible runs.
--no-seed	Run unseeded even when the graph file has a settings.seed.
--deterministic	Ask PyTorch for deterministic kernels (warn_only, so an op with no deterministic implementation warns rather than failing the run).

# Validate an architecture without running it
python run_graph.py ../examples/Model_Architecture/ResNet-SkipConnection-CNN/graph.json --validate-only

Exit codes

Code	Meaning
0	The graph ran, or it is valid with --validate-only
1	The file is missing, the graph fails validation, or a node fails at runtime (-v adds the full traceback)
2	Bad command line (argument parsing)

Plugin nodes

The runner loads the plugin packs listed in the plugin lockfile under CODEFYUI_USER_DATA_DIR, or, when that variable is unset, under the platform data directory (%LOCALAPPDATA%\codefyui, ~/.local/share/codefyui or ~/Library/Application Support/codefyui). The cdui commands, cdui plugin install and servers started with cdui start included, keep their lockfile in <install dir>/.codefyui_dev/ instead. To run a graph that uses nodes from those packs, set CODEFYUI_USER_DATA_DIR=<install dir>/.codefyui_dev first; otherwise the runner does not know those node types and validation fails.

Relative file paths

A relative path in a reader node is resolved by the process that runs the graph. For the runner, the working directory is the one you start it in (backend/ in the commands above).

• ImageReader uses an absolute path as written. A relative path is looked up first in the image upload store (backend/data/images by default), then relative to the working directory.
• CSVReader looks a bare file name up first in the data-file upload store (backend/data/files by default). Otherwise a relative path resolves against the working directory; with CODEFYUI_PROJECT_DIR set, it resolves inside that project directory instead and may not leave it (the bundled data/samples/iris.csv is the exception).

Where graphs come from

Any graph exported from the UI (Tabs & Persistence → Import / export) is a plain JSON file in the same format, so you can build a pipeline visually and then run it from the CLI. The bundled examples under examples/ are ready to run — see the Examples Gallery.