Feature

Code execution without the risk.

Agents execute scripts in isolated sandboxes with no access to production and no side effects. Run data analysis, generate reports, and transform files, all inside a controlled environment.

What you get

Script execution

Agents can write and execute scripts as part of their runs. Data analysis, file processing, API calls - if you can describe it, an agent can run it.

Isolated environment

Scripts run in isolated sandboxes with no access to your production systems. The sandbox is provisioned per-run and torn down when the run completes.

Multiple runtimes

Support for common scripting languages. Agents pick the right tool for the job and execute code in a controlled environment.

Captured outputs

Everything a script produces (console output, error logs, generated files, and status codes) is captured and stored as part of the run log.

Persistent filesystem access

Sandboxes can read from and write to the project filesystem. Scripts can process files from previous runs or save results for future ones.

Agent-driven

The agent decides when to run code, what to run, and how to interpret the results. Scripts are a tool in the agent's toolkit, not a separate workflow.

How it works

Enable the execute_code permission on an agent to let it run scripts during its runs.

During execution, the agent writes code and submits it to the sandbox. The platform provisions an isolated environment.

The script runs with access to the project filesystem but no external network or production systems.

Outputs are captured (console output, files, status codes) and the agent uses them to continue its work or produce artifacts.

Ready to get started?

Set up your first project in minutes. Connect your tools, configure an agent, and let it run.