FAQ¶

The MolCraftDiff CLI could not be found in the current Python environment. Run automolcraft doctor — it shows which Python the app uses and whether MolCraftDiffusion is installed there. Install it following Installation, then restart the backend.

automolcraft serve always uses the environment it is installed in. With ./dev.sh, if you installed into a different environment than the one dev.sh picks, set the interpreter explicitly:

BACKEND_PYTHON=/path/to/your/python ./dev.sh

No models appear in the generation tab¶

The app could not find any checkpoint folders in MOLCRAFT_MODELS_DIR. Each checkpoint folder must contain edm_chem.pkl.

Check the current models directory shown under the model list. If it is wrong, set MOLCRAFT_MODELS_DIR in your .env file and restart.

xTB calculations fail immediately¶

xtb is not on the PATH used by the backend (automolcraft doctor shows it under Analysis tools). Install it via conda-forge in the same environment:

conda install -c conda-forge --override-channels xtb==6.7.1 xtb-python

xtb-python is needed by XTB electronic properties; the xtb binary alone covers geometry optimisation.

Then restart the backend.

Port conflict on startup¶

If port 8000 is already in use:

automolcraft serve --port 9000
# or, with the manual install:
BACKEND_PORT=9000 ./dev.sh

Then open http://localhost:9000.

The frontend shows a blank page or "cannot connect"¶

The pre-built frontend in frontend/dist may be missing or stale (for example after git pull). Rebuild it and start again:

automolcraft serve --rebuild

With the manual install, run npm install && npm run build in webapp/database-explorer-lite/frontend, then restart ./dev.sh. To run Vite directly at :5173 instead, use automolcraft serve --dev (or FRONTEND_DEV=1 ./dev.sh).

An analysis job completes but "Apply results" produces no new columns¶

The runner may have returned an empty result (e.g. all molecules timed out). Check the job log by clicking the terminal icon next to the job in the queue. Common causes:

Timeout too short for the dataset size — increase Timeout per molecule.
Molecule converter failed to assign bonds — switch Molecule converter from xyz2mol to rdkit (or vice versa).
xTB method incompatible with element types in the dataset — try a different XTB method.

Dataset is loaded but Visualization shows no plots¶

Panels must be added manually. In the Visualization action bar, click 2D Scatter, 3D Scatter, or Histogram to add a plot panel, then click its axis labels to change the column bindings.

`ValueError: numpy.dtype size changed` after installing analysis packages¶

pandas was downgraded to a build made for numpy 1. The usual cause is pip install 'molcraftdiffusion[analyze]', which pulls in posecheck and its pandas==2.0.0 pin. Reinstall a numpy-2 pandas:

pip install 'pandas>=2.3,<2.4'

pip will then warn that posecheck wants pandas 2.0.0; that only affects protein-pocket checks. The environment files avoid this by not installing posecheck.

Where do xTB's scratch files go?¶

xTB writes files such as xtbopt.log, charges, wbo and xtbrestart into its working directory. Analysis and generation jobs run inside their own job folder (under MOLCRAFT_ANALYSIS_WORK_DIR or MOLCRAFT_OUTPUTS_DIR), so those files stay with the job. If you find them in the repository root, they come from an older version of the app and can be deleted.

On macOS, the health check says the MolCraftDiffusion commit does not match¶

Expected. environment-macos.yml installs MolCraftDiffusion 1.12.0 at the commit that added macOS support, which is newer than the Linux pin. The version still matches. See macOS (Apple Silicon).