Reflex Docs
Guides

FAQ

Common questions about the CLI and SDK.

Can I use the SDK without the CLI?

Yes. Set REFLEX_API_KEY and the SDK is fully self-contained. The CLI is the recommended way to obtain that key (reflex login), but any key minted from the dashboard works the same way.

Can I run inference offline?

No. Inference runs on hosted GPUs. The control loop runs locally on your robot machine, but every action chunk requires a round-trip to the server.

What's the difference between dry_run and apply_actions?

  • dry_run exchanges observations and action chunks but does not enable hardware motion. Use it for development and smoke testing.
  • apply_actions actually drives the robot. Validate the deployment with reflex doctor before enabling motion.

Mode is set in your YAML config (mode: dry_run / mode: apply_actions); an existing session can be moved between modes with reflex sessions promote.

How do I bring my own robot?

Two options:

  1. Use the subprocess hardware connector in your YAML config to run shell commands that read your hardware and print/consume the JSON shapes documented in Observation and action schema.
  2. Subclass reflex.connectors.base.RobotConnector in Python and register it via the YAML hardware.kind field. See the SDK overview for the connector ABC.

How do I bring my own model?

Train a LoRA adapter against pi0.5 with reflex training start, then deploy the resulting artifact and drive it with reflex connect:

reflex deploy --artifact model_abc123 --name my-deployment
reflex connect --config robot.yaml

Full fine-tunes are also supported (--trainable full). To ship a fully custom policy instead, see reflex model.

How do I run a CI-friendly check?

Validate a deployment's readiness, machine-readable, without touching hardware:

reflex doctor <deployment-id> --json

For a bring-your-own-model project, reflex model test runs your Policy's predict() on a synthetic observation with no GPU.

Are there other-language SDKs?

The Python SDK is the supported surface. Other languages are not currently shipped; if you need them, drive the CLI from your language of choice — its --json output is stable and scriptable.

Where do I report bugs or request features?

GitHub issues: github.com/reflex-inc/reflex.

Error reference

Common failures from the CLI and SDK, what they mean, and how to fix them.

CLI overview

Install the Reflex CLI, learn the global flags, and find the right command group.

On this page

Can I use the SDK without the CLI?Can I run inference offline?What's the difference between dry_run and apply_actions?How do I bring my own robot?How do I bring my own model?How do I run a CI-friendly check?Are there other-language SDKs?Where do I report bugs or request features?