Reflex Docs
CLI

reflex model

Scaffold, test, and deploy your own Policy.

reflex model is the workflow for bringing your own model. A Policy is a small Python class with load() and predict(); these commands scaffold one, test it locally, and produce a deploy manifest. The flow is model-agnostic.

reflex model init

Scaffold a Policy stub plus a reflex.model.json that runs out of the box.

reflex model init mylab/grasp-v1
FlagDefaultDescription
NAMEPolicy name (positional argument), e.g. mylab/grasp-v1.
--dir, -d.Target directory.

This writes policy.py (edit load() / predict()) and reflex.model.json.

reflex model test

Run the Policy's predict() locally on a synthetic observation — no GPU required. Prints the returned action-chunk shape.

reflex model test
FlagDefaultDescription
--dir, -d.Model directory.

reflex model deploy

Validate the Policy and emit a deploy manifest for the platform.

reflex model deploy
FlagDefaultDescription
--dir, -d.Model directory.