Console UI

The Console is the human projection of the same project resources used by the SDK and MCP tools. Use it where visual density or accountable judgment is the work. Keep repeatable integration and automation in Python.

Open the sx Console

Sign in, select the same project UUID used by your SDK client, and open the workspace that owns the decision in front of you.

Where the UI earns its place

Workspace	Open it to	Keep in SDK or MCP
Library	browse episodes, search visually, inspect lineage and coverage	repeatable queries and snapshot creation
Capture	inspect stations and live capture campaigns	programmatic capture submission and monitoring
Label	review frames, edit subtasks, resolve annotations	bulk selection and downstream data automation
Training	watch live metrics, compare runs, read reports	training submission and experiment programs
Improve	steer a durable agent campaign and decide stop/continue	bounded agent execution and evidence gathering
Fleet	commission hardware, watch live state, handle escalations, supervise motion	deployment automation and status integrations
Settings	manage projects, grants, API keys, sessions, and organization context	runtime use of already-issued credentials

Worlds does not need a separate workspace. Evaluation evidence appears in the workflows that use it, with deep spatial and temporal inspection linked to Rerun.

One object across three surfaces

Python SDK                       MCP + Skill                 Console
client.worlds.evaluate(...)  →   worlds.evaluate         →  evaluation evidence
operation.watch()            →   operation.updates       →  live status
WorldEvaluation              →   exact evidence refs     →  human-readable inspection

The Console does not infer success from a button click or agent message. Durable work is complete only when the owning service records a terminal result.

A practical handoff

Build or submit in Python

Start repeatable work through the SDK and persist the operation reference.

Delegate routine monitoring

Let an agent follow bounded updates, assemble evidence, and surface refusals.

Open the Console for the decision

Compare visual evidence, inspect the affected resource, and make the accountable label, stop/continue, escalation, or promotion decision.

Continue from the same identity

Feed the exact resulting reference—not a screenshot or copied display value—into the next SDK call.

A UI is not required for ordinary automation

A coding user can operate the platform entirely through the SDK and agent plugin. Open the Console when seeing or deciding is materially better than expressing the same action in code.