# Dreamscale

> Cloud robot policy inference through a small Python SDK and CLI.

> ## Agent Instructions
> Treat the Dropbear Python SDK and CLI as the supported public integration surfaces.
> Prefer context-managed sessions and model-specific observation helpers.
> Do not infer or recommend internal control-plane APIs.

- [Dropbear](https://docs.dreamscalelabs.com/index.md): Run robot policies in the cloud through a small Python SDK and CLI.
- [Python quickstart](https://docs.dreamscalelabs.com/quickstart.md): Send a real robot observation to Dropbear and inspect one action chunk without actuating.
- [Installation](https://docs.dreamscalelabs.com/installation.md): Install the Dropbear SDK and CLI, authenticate, and keep the launch package up to date.
- [Python SDK](https://docs.dreamscalelabs.com/sdk/overview.md): Own the Dropbear policy lifecycle from connection through prediction, action buffering, and deterministic cleanup.
- [Model contracts](https://docs.dreamscalelabs.com/sdk/model-contracts.md): Match camera, state, action ordering, units, chunk size, and rate for the live Dropbear models.
- [SO-101](https://docs.dreamscalelabs.com/guides/so101.md): Configure, calibrate, diagnose, dry-run, and safely confirm motion on an SO-101.
- [Franka inference](https://docs.dreamscalelabs.com/guides/franka.md): Build a MolmoAct2-DROID observation from a Franka stack and inspect one action chunk without actuating.
- [Inspect Robots](https://docs.dreamscalelabs.com/guides/inspect-robots.md): Run a Dropbear-hosted policy on a physical robot from an existing Inspect Robots evaluation, using DreamZero-YAM on a bimanual YAM as the worked example.
- [CLI reference](https://docs.dreamscalelabs.com/reference/cli.md): Use the supported Dropbear command tree for authentication, diagnostics, sessions, simulation, and SO-101 setup.
- [Troubleshooting](https://docs.dreamscalelabs.com/troubleshooting.md): Recover from common Dropbear authentication, capacity, transport, session, camera, calibration, and contract failures.