sx
Build robot systems with sx
SDK-first documentation for data, simulation, training, deployment, and agentic operations
Build through the typed Python SDK. Delegate bounded work through MCP + Skills. Open the Console when a human needs to inspect evidence, steer an agent, label data, or supervise a fleet.
Start from code
from uuid import UUID
import sx
async def inspect_task(project_id: UUID) -> None:
async with sx.connect(project_id) as client:
task = await client.experience.task("task:pick-mug@production")
print(task)sx.connect(...) binds one explicit project and exposes four typed doors:
client.experience, client.worlds, client.autonomy, and client.fleet. There are no service
URLs or request dictionaries on the ordinary path.
Experience
Resolve tasks and embodiments, capture episodes, search the corpus, and freeze immutable training snapshots.
Worlds
Generate experience, open bounded sessions, and evaluate exact applications.
Autonomy
Train application manifests, approve passing evidence, and run bounded improvement loops.
Fleet
Deploy complete applications, run supervised research, and restore a healthy release.
The same contract everywhere
| Surface | Best for | What comes back |
|---|---|---|
| Python SDK | products, notebooks, pipelines, tests, and automation | canonical Python values and typed operation handles |
| MCP + Skills | investigation and bounded multi-step work | the same resource and operation references, summarized with evidence |
| Console UI | visual inspection and accountable human decisions | views and controls projected from the same platform state |
Long-running work is durable on every surface. Save its operation reference, disconnect safely, and reattach later. Submission is not success; only a terminal result is.
The shortest useful path
Install the SDK, authenticate once, connect one explicit project, resolve one exact resource, then run one durable operation and save its reference.
Choose your next step
- New to the platform: install and make your first SDK call
- Building data pipelines: use the Experience SDK
- Evaluating a robot application: use the Worlds SDK
- Training or promoting a candidate: use the Autonomy SDK
- Adding an agent to your workflow: install MCP + Skills