Product / Sandboxed execution
Sandboxed execution
Agents and long jobs run in containers with no network, a read-only file system and resource limits — inside your boundary.
For IT and developers who run other people's code next to their data.
- Available Run code inside your boundary
Key features
- No network
- A run cannot reach the network or a registry.
- Read-only root
- The container's file system cannot be changed.
- Resource limits
- Memory, time, CPU, processes, output size and log size are bounded.
- Explicit grants
- Running shared work needs a member role and an execution grant.
- Recorded
- Each run records its inputs and outcome.
- Your hardware
- Runs on your servers; nothing is sent to a hosted service.
How it works
- A person or an approved agent starts a run.
- The runner starts one container for the job with its limits.
- Outputs are checked against their limits before they are kept.
Technical details
The limits per job, by default, are set in the runner itself:
DEFAULTS = {'memory_mb': 256, 'seconds': 60, 'cpus': 1.0, 'pids': 64Not yet, and not planned
- Run code inside your boundary: No GPU or distributed scheduling
Not planned: suite-wide petrel or techlog replacement; a general application builder or marketplace; universal format conversion.