Code and agents
Read, run, and verify code beside the task.
The built-in code sandbox identifies the resource type and shows source beside the result. The agent works inside the selected project boundary, attaches tool events and results to the run, and never replaces real data with demo responses.

- Editor and live preview for supported code
- Personal and project agents
- Links to files, tasks, reports, and GitHub
Getting started
Four steps to a working scenario.
Every step uses real Cabinet surfaces and preserves project access rules.
- 01
Add code
Create an HTML resource or open a supported file from storage.
- 02
Inspect the result
The editor and isolated preview make changes visible immediately.
- 03
Brief an agent
Select the project context, skills, model, and permitted tools.
- 04
Receive a trace
The run stores events, tool output, source links, and errors.