> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dreamscalelabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Treat the Dreamscale 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.

# Dreamscale

> Run robot policies in the cloud through a small Python SDK and CLI.

# Cloud inference for your robot

Dreamscale runs robot policies in the cloud while your local controller keeps
ownership of sensing, actuation, and safety.

```mermaid theme={null}
flowchart LR
  A["Robot cameras and state"] --> B["Dreamscale Python SDK"]
  B --> C["Cloud policy"]
  C --> D["Action chunk"]
  D --> E["Your validated robot controller"]
```

<CardGroup cols={2}>
  <Card title="Integrate with Python" icon="code" href="/quickstart">
    Send one real robot observation and inspect the returned action chunk
    without moving the robot.
  </Card>

  <Card title="Set up an SO-101" icon="robot" href="/guides/so101">
    Configure, calibrate, diagnose, dry-run, and safely start the supported arm
    workflow.
  </Card>
</CardGroup>

## What Dreamscale owns

* authenticated cloud sessions;
* policy worker allocation and startup;
* observation transport and action-chunk delivery;
* automatic QUIC-to-relay transport selection;
* session lifecycle and usage reporting.

## What you own

* cameras and robot state;
* conversion into the selected model's observation contract;
* action validation and robot-coordinate conversion;
* local watchdogs, limits, collision handling, and e-stop behavior;
* deterministic session cleanup.

<Note>
  Open sessions consume prepaid credits. Use a context manager or call
  `close()` when work is complete. View current availability, balance, and
  usage in the
  [dashboard](https://app.dreamscalelabs.com/dashboard).
</Note>

## Choose your next step

* Have a controller already? Start with the [Python quickstart](/quickstart).
* Need package or platform details? Read [Installation](/installation).
* Need exact shapes and units? Open [Model contracts](/sdk/model-contracts).
* Something failed? Use [Troubleshooting](/troubleshooting).